Add files using upload-large-folder tool

data/README.md

@@ -0,0 +1,14 @@
+This model repository does not upload the AlphaFold3 dataset payload.
+
+Download the required dataset separately:
+
+```bash
+modelscope download --dataset OneScience/AlphaFold3_dataset
+```
+
+Then link the dataset package data directory into this model package:
+
+```bash
+mkdir -p data
+ln -s /path/to/AlphaFold3_dataset/data data/alphafold3_dataset
+```

flax_model/alphafold3/_tools/hmmer/share/man/man1/alimask.1

@@ -0,0 +1,362 @@
+.TH "alimask" 1 "Aug 2023" "HMMER 3.4" "HMMER Manual"
+
+.SH NAME
+alimask \- calculate and add column mask to a multiple sequence alignment
+
+.SH SYNOPSIS
+.B alimask
+[\fIoptions\fR]
+.I msafile
+.I postmsafile
+
+
+.SH DESCRIPTION
+
+.PP
+.B alimask
+is used to apply a mask line to a multiple sequence
+alignment, based on provided alignment or model coordinates.
+When 
+.B hmmbuild
+receives a masked alignment as input, it produces a profile
+model in which the emission probabilities at masked positions
+are set to match the background frequency, rather than being
+set based on observed frequencies in the alignment. 
+Position-specific insertion and deletion rates are not 
+altered, even in masked regions. 
+.B alimask
+autodetects input format, and produces masked alignments 
+in Stockholm format.  
+.I msafile 
+may contain only one sequence alignment.
+
+.PP
+A common motivation for masking a region in an alignment is
+that the region contains a simple tandem repeat that is 
+observed to cause an unacceptably high rate of false positive
+hits. 
+
+.PP
+In the simplest case, a mask range is given in coordinates 
+relative to the input alignment, using
+.BI \-\-alirange " <s>". 
+However it is more often the case that the region to be 
+masked has been identified in coordinates relative to 
+the profile model (e.g. based on recognizing a simple 
+repeat pattern in false hit alignments or in the HMM logo). 
+Not all alignment columns are converted to match state 
+positions in the profile (see the 
+.B \-\-symfrac
+flag for 
+.B hmmbuild
+for discussion), so model positions do not necessarily match 
+up to alignment column positions. 
+To remove the burden of converting model positions to 
+alignment positions, 
+.B alimask
+accepts the mask range input in model coordinates as well, 
+using
+.BI \-\-modelrange " <s>". 
+When using this flag, 
+.I alimask
+determines which alignment positions would be identified by
+.B hmmbuild
+as match states, a process that requires that all
+.B hmmbuild 
+flags impacting that decision be supplied to 
+.BR alimask .
+It is for this reason that many of the 
+.B hmmbuild 
+flags are also used by
+.BR alimask .  
+
+
+
+.SH OPTIONS
+
+.TP
+.B \-h
+Help; print a brief reminder of command line usage and all available
+options.
+
+.TP
+.BI \-o " <f>"
+Direct the summary output to file
+.IR <f> ,
+rather than to stdout.
+
+
+.SH OPTIONS FOR SPECIFYING MASK RANGE
+
+A single mask range is given as a dash-separated pair, like
+.B "\-\-modelrange 10\-20"
+and multiple ranges may be submitted as a comma-separated list,
+.BR "\-\-modelrange 10\-20,30\-42" .
+
+
+.TP
+.BI \-\-modelrange " <s>"
+Supply the given range(s) in model coordinates. 
+
+.TP
+.BI \-\-alirange " <s>"
+Supply the given range(s) in alignment coordinates. 
+
+.TP
+.B \-\-appendmask 
+Add to the existing mask found with the alignment.
+The default is to overwrite any existing mask. 
+
+.TP
+.BI \-\-model2ali " <s>"
+Print model range(s) and the corresponding alignment range(s).
+No masked alignment is produced.
+The output is a single line for each input range, of the form
+.nf
+   i..j -> m..n
+.fi
+with i & j representing model range values, and m & n
+representing alignment range values.
+
+.TP
+.BI \-\-ali2model " <s>"
+Print alignment range(s) and the corresponding model range(s).
+No masked alignment is produced.
+Because some alignment positions may not map to model positions,
+the range(s) produced will begin with the first alignment
+position between <from> and <to> (inclusive) that maps to a
+model position, and end with the final alignment position in
+that range that maps to a model position.
+The output is a single line for each input range, of
+the form
+.nf
+  i..j -> m..n
+.fi
+with i & j representing alignment range values, and m & n
+representing model range values.
+If no alignment positions in the range <from>..<to> map
+to a model position, the output prints the input <from>
+and <to> mapping to nothing, with the format:
+.nf
+ i..j  ->   \-..\-  (no map) .
+.fi
+
+
+.SH OPTIONS FOR SPECIFYING THE ALPHABET
+
+.TP
+.B \-\-amino
+Assert that sequences in 
+.I msafile
+are protein, bypassing alphabet autodetection.
+
+.TP
+.B \-\-dna
+Assert that sequences in
+.I msafile
+are DNA, bypassing alphabet autodetection.
+
+.TP
+.B \-\-rna
+Assert that sequences in 
+.I msafile
+are RNA, bypassing alphabet autodetection.
+
+
+
+.SH OPTIONS CONTROLLING PROFILE CONSTRUCTION 
+
+These options control how consensus columns are defined in an alignment.
+
+.TP
+.B \-\-fast 
+Define consensus columns as those that have a fraction >= 
+.B symfrac
+of residues as opposed to gaps. (See below for the
+.B \-\-symfrac
+option.) This is the default.
+
+.TP
+.B \-\-hand
+Define consensus columns in next profile using reference annotation to
+the multiple alignment. 
+This allows you to define any consensus columns you like.
+
+.TP
+.BI \-\-symfrac " <x>"
+Define the residue fraction threshold necessary to define a
+consensus column when using the 
+.B \-\-fast 
+option. The default is 0.5. The symbol fraction in each column
+is calculated after taking relative sequence weighting into account,
+and ignoring gap characters corresponding to ends of sequence
+fragments
+(as opposed to internal insertions/deletions).
+Setting this to 0.0 means that every alignment column will be assigned
+as consensus, which may be useful in some cases. Setting it to 1.0
+means that only columns that include 0 gaps (internal
+insertions/deletions) will be assigned as consensus.
+
+.TP
+.BI \-\-fragthresh " <x>"
+We only want to count terminal gaps as deletions if the aligned
+sequence is known to be full-length, not if it is a fragment (for
+instance, because only part of it was sequenced). HMMER uses a simple
+rule to infer fragments: if the sequence length L is less than 
+or equal to a fraction
+.I <x> 
+times the alignment length in columns,
+then the sequence is handled as a fragment. The default is 0.5.
+Setting
+.B \-\-fragthresh 0
+will define no (nonempty) sequence as a fragment; you might want to do
+this if you know you've got a carefully curated alignment of full-length
+sequences.
+Setting
+.B \-\-fragthresh 1
+will define all sequences as fragments; you might want to do this if
+you know your alignment is entirely composed of fragments, such as
+translated short reads in metagenomic shotgun data.
+
+
+.SH OPTIONS CONTROLLING RELATIVE WEIGHTS
+
+HMMER uses an ad hoc sequence weighting algorithm to downweight
+closely related sequences and upweight distantly related ones. This
+has the effect of making models less biased by uneven phylogenetic
+representation. For example, two identical sequences would typically
+each receive half the weight that one sequence would.  These options
+control which algorithm gets used.
+
+.TP
+.B \-\-wpb
+Use the Henikoff position-based sequence weighting scheme [Henikoff
+and Henikoff, J. Mol. Biol. 243:574, 1994].  This is the default.
+
+.TP 
+.B \-\-wgsc 
+Use the Gerstein/Sonnhammer/Chothia weighting algorithm [Gerstein et
+al, J. Mol. Biol. 235:1067, 1994].
+
+.TP 
+.B \-\-wblosum
+Use the same clustering scheme that was used to weight data in
+calculating BLOSUM substitution matrices [Henikoff and Henikoff,
+Proc. Natl. Acad. Sci 89:10915, 1992]. Sequences are single-linkage
+clustered at an identity threshold (default 0.62; see
+.BR \-\-wid )
+and within each cluster of c sequences, each sequence gets relative
+weight 1/c.
+
+.TP
+.B \-\-wnone
+No relative weights. All sequences are assigned uniform weight. 
+
+.TP 
+.BI \-\-wid " <x>"
+Sets the identity threshold used by single-linkage clustering when 
+using 
+.BR \-\-wblosum . 
+Invalid with any other weighting scheme. Default is 0.62.
+
+
+
+
+
+.SH OTHER OPTIONS
+
+.TP 
+.BI \-\-informat " <s>"
+Assert that input
+.I msafile
+is in alignment format
+.IR <s> ,
+bypassing format autodetection.
+Common choices for 
+.I <s> 
+include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+For more information, and for codes for some less common formats,
+see main documentation.
+The string
+.I <s>
+is case-insensitive (\fBa2m\fR or \fBA2M\fR both work).
+
+
+.TP
+.BI \-\-outformat " <s>"
+Write the output
+.I postmsafile
+in alignment format
+.IR <s> .
+Common choices for 
+.I <s> 
+include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+The string
+.I <s>
+is case-insensitive (\fBa2m\fR or \fBA2M\fR both work).
+Default is
+.BR stockholm .
+
+
+.TP 
+.BI \-\-seed " <n>"
+Seed the random number generator with
+.IR <n> ,
+an integer >= 0. 
+If 
+.I <n> 
+is nonzero, any stochastic simulations will be reproducible; the same
+command will give the same results.
+If 
+.I <n>
+is 0, the random number generator is seeded arbitrarily, and
+stochastic simulations will vary from run to run of the same command.
+The default seed is 42.
+
+
+
+.SH SEE ALSO 
+
+See 
+.BR hmmer (1)
+for a master man page with a list of all the individual man pages
+for programs in the HMMER package.
+
+.PP
+For complete documentation, see the user guide that came with your
+HMMER distribution (Userguide.pdf); or see the HMMER web page
+(http://hmmer.org/).
+
+
+
+.SH COPYRIGHT
+
+.nf
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi
+
+For additional information on copyright and licensing, see the file
+called COPYRIGHT in your HMMER source distribution, or see the HMMER
+web page 
+(http://hmmer.org/).
+
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi
+

flax_model/alphafold3/_tools/hmmer/share/man/man1/esl-afetch.1

@@ -0,0 +1,133 @@
+.TH "esl\-afetch" 1 "Aug 2023" "Easel 0.49" "Easel Manual"
+
+.SH NAME
+esl\-afetch \- retrieve alignments from a multi-MSA database
+
+.SH SYNOPSIS
+
+.nf
+\fBesl\-afetch\fR [\fIoptions\fR] \fImsafile key\fR
+  (single MSA retrieval)
+
+\fBesl\-afetch \-f\fR [\fIoptions\fR] \fImsafile keyfile\fR
+  (multiple MSA retrieval, from a file of keys)
+
+\fBesl\-afetch \-\-index \fR\fImsafile\fR
+  (index an MSA file for retrieval)
+
+.SH DESCRIPTION
+
+.B esl\-afetch
+retrieves the alignment named
+.I key
+from an alignment database in file
+.I msafile.
+The
+.I msafile
+is a "multiple multiple alignment" file in Stockholm (e.g. native
+Pfam or Rfam) format.
+The 
+.I key
+is either the name (ID) of the alignment, or its accession
+number (AC).
+
+.PP
+Alternatively, 
+.B esl\-afetch \-f
+provides the ability to fetch many alignments at once.
+The 
+.B \-f 
+option has it interpret the second argument as a
+.IR keyfile ,
+a file consisting of one name or accession per line.
+
+.PP
+The
+.I msafile
+should first be SSI indexed with
+.B esl\-afetch \-\-index
+for efficient retrieval. An SSI index is
+not required, but without one alignment retrieval may
+be painfully slow.
+
+.SH OPTIONS
+
+.TP
+.B \-h
+Print brief help; includes version number and summary of
+all options, including expert options.
+
+.TP
+.B \-f
+Interpret the second argument as a 
+.I keyfile
+instead of as just one
+.IR key . 
+The
+.I keyfile
+contains one name or accession per line.
+This option doesn't work with the
+.B \-\-index
+option.
+
+
+.TP
+.BI \-o " <f>"
+Output retrieved alignments to a file 
+.I <f>
+instead of to stdout.
+
+.TP
+.BI \-O
+Output retrieved alignment to a file named
+.IR key .
+This is a convenience for saving some typing:
+instead of 
+.nf
+ \fB% esl\-afetch \-o RRM_1 msafile RRM_1\fR
+.fi
+you can just type
+.nf
+ \fB% esl\-afetch \-O msafile RRM_1\fR
+.fi
+The
+.B \-O 
+option only works if you're retrieving a
+single alignment; it is incompatible with 
+.B \-f.
+
+.TP
+.B \-\-index
+Instead of retrieving a
+.I key,
+the special command
+.B esl\-afetch \-\-index
+.I msafile
+produces an SSI index of the names and accessions
+of the alignments in
+the 
+.I msafile.
+Indexing should be done once on the
+.I msafile
+to prepare it for all future fetches.
+
+
+.SH SEE ALSO
+
+.nf
+http://bioeasel.org/
+.fi
+
+.SH COPYRIGHT
+
+.nf 
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi 
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi
+

flax_model/alphafold3/_tools/hmmer/share/man/man1/esl-alimanip.1

@@ -0,0 +1,383 @@
+.TH "esl\-alimanip" 1  "Aug 2023" "Easel 0.49" "Easel Manual"
+
+.SH NAME
+esl\-alimanip \- manipulate a multiple sequence alignment
+
+.SH SYNOPSIS
+
+.B esl\-alimanip
+[\fIoptions\fR]
+.I msafile
+
+.SH DESCRIPTION
+
+.B esl\-alimanip
+can manipulate the multiple sequence alignment(s) in 
+.I msafile
+in various ways. Options exist to remove
+specific sequences, reorder sequences, designate reference columns
+using Stockholm "#=GC RF" markup, and add annotation that numbers
+columns. 
+
+.PP
+The alignments can be of protein or DNA/RNA sequences. All alignments
+in the same 
+.I msafile
+must be either protein or DNA/RNA. The alphabet will be autodetected
+unless one of the options 
+.B \-\-amino,
+.B \-\-dna,
+or 
+.B \-\-rna 
+are given. 
+
+
+
+.SH OPTIONS
+
+.TP
+.B \-h 
+Print brief help;  includes version number and summary of
+all options, including expert options.
+
+.TP
+.BI \-o " <f>"
+Save the resulting, modified alignment in Stockholm format to a file
+.I <f>.
+The default is to write it to standard output.
+
+.TP 
+.BI \-\-informat " <s>"
+Assert that 
+.I msafile
+is in alignment format
+.IR <s> ,
+bypassing format autodetection.
+Common choices for 
+.I <s> 
+include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+For more information, and for codes for some less common formats,
+see main documentation.
+The string
+.I <s>
+is case-insensitive (\fBa2m\fR or \fBA2M\fR both work).
+
+
+.TP 
+.BI \-\-outformat " <s>"
+Write the output in alignment format
+.IR <s> .
+Common choices for 
+.I <s> 
+include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+The string
+.I <s>
+is case-insensitive (\fBa2m\fR or \fBA2M\fR both work).
+Default is
+.BR stockholm .
+
+.TP
+.B \-\-devhelp
+Print help, as with  
+.B \-h,
+but also include undocumented developer options. These options are not
+listed below, are under development or experimental, and are not
+guaranteed to even work correctly. Use developer options at your own
+risk. The only resources for understanding what they actually do are
+the brief one-line description printed when
+.B \-\-devhelp
+is enabled, and the source code.
+
+.SH EXPERT OPTIONS
+
+.TP 
+.BI \-\-lnfract " <x>"
+Remove any sequences with length less than 
+.I <x>
+fraction the length of the median length sequence in the alignment.
+
+.TP 
+.BI \-\-lxfract " <x>"
+Remove any sequences with length more than 
+.I <x>
+fraction the length of the median length sequence in the alignment.
+
+.TP 
+.BI \-\-lmin " <n>"
+Remove any sequences with length less than 
+.I <n>
+residues.
+
+.TP 
+.BI \-\-lmax " <n>"
+Remove any sequences with length more than 
+.I <n>
+residues.
+
+.TP 
+.BI \-\-rfnfract " <x>"
+Remove any sequences with nongap RF length less than 
+.I <x>
+fraction the nongap RF length of the alignment.
+
+.TP 
+.BI \-\-detrunc " <n>"
+Remove any sequences that have all gaps in the first 
+.I <n>
+non-gap #=GC RF columns or the last 
+.I <n>
+non-gap #=GC RF columns.
+
+.TP 
+.BI \-\-xambig " <n>"
+Remove any sequences that has more than
+.I <n>
+ambiguous (degenerate) residues.
+
+.TP 
+.BI \-\-seq\-r " <f>"
+Remove any sequences with names listed in file 
+.I <f>.
+Sequence names listed in 
+.I <f>
+can be separated by tabs, new lines, or spaces.
+The file must be in Stockholm format for this option to work. 
+
+.TP 
+.BI \-\-seq\-k " <f>"
+Keep only sequences with names listed in file 
+.I <f>.
+Sequence names listed in 
+.I <f>
+can be separated by tabs, new lines, or spaces.
+By default, the kept sequences will remain in the original order
+they appeared in 
+.I msafile,
+but the order from 
+.I <f> 
+will be used if the 
+.B \-\-k\-reorder
+option is enabled.
+The file must be in Stockholm format for this option to work. 
+
+.TP 
+.B \-\-small
+With
+.B \-\-seq\-k 
+or
+.B \-\-seq\-r,
+operate in small memory mode. 
+The alignment(s) will not be stored in memory, thus
+.B \-\-seq\-k 
+and
+.B \-\-seq\-r
+will be able to work on very large alignments regardless
+of the amount of available RAM.
+The alignment file must be in Pfam
+format and 
+.B \-\-informat pfam
+and one of
+.B \-\-amino,
+.B \-\-dna,
+or
+.B \-\-rna
+must be given as well.
+
+.TP 
+.B \-\-k\-reorder
+With
+.BI \-\-seq\-k " <f>",
+reorder the kept sequences in the output alignment to the order
+from the list file
+.I <f>.
+
+.TP 
+.BI \-\-seq\-ins " <n>"
+Keep only sequences that have at least 1 inserted residue after 
+nongap RF position 
+.I <n>.
+
+.TP 
+.BI \-\-seq\-ni " <n>"
+With 
+.B \-\-seq\-ins
+require at least 
+.I <n> 
+inserted residues in a sequence for it to be kept.
+
+.TP 
+.BI \-\-seq\-xi " <n>"
+With 
+.B \-\-seq\-ins
+allow at most
+.I <n> 
+inserted residues in a sequence for it to be kept.
+
+.TP 
+.BI \-\-trim " <f>"
+File 
+.I <f>
+is an unaligned FASTA file containing truncated versions of each
+sequence in the 
+.I msafile. 
+Trim the sequences in the alignment to match their truncated versions
+in 
+.I <f>.
+If the alignment output format is Stockholm (the default output
+format), all per-column (GC) and per-residue (GR) annotation will be
+removed from the alignment when
+.B \-\-trim
+is used. However, if 
+.B \-\-t\-keeprf 
+is also used, the reference annotation (GC RF) will be kept.
+
+.TP 
+.B \-\-t\-keeprf
+Specify that the 'trimmed' alignment maintain the original
+reference (GC RF) annotation. Only works in combination with 
+.B \-\-trim.
+
+.TP 
+.BI \-\-minpp " <x>"
+Replace all residues in the alignments for which the posterior
+probability annotation (#=GR PP) is less than 
+.I <x>
+with gaps. The PP annotation for these residues is also converted to
+gaps. 
+.I <x>
+must be greater than 0.0 and less than or equal to 0.95.
+
+.TP 
+.BI \-\-tree " <f>"
+Reorder sequences by tree order. 
+Perform single linkage clustering on the sequences in the alignment
+based on sequence identity given the alignment to define a 'tree' 
+of the sequences. The sequences in the alignment are reordered
+according to the tree, which groups similar sequences together. The
+tree is output in Newick format to 
+.I <f>.
+
+.TP 
+.BI \-\-reorder " <f>"
+Reorder sequences to the order listed in file 
+.I <f>.
+Each sequence in the alignment must be listed in 
+.I <f>.
+Use
+.B \-\-k\-reorder
+to reorder only a subset of sequences to a subset alignment file. 
+The file must be in Stockholm format for this option to work. 
+
+.TP 
+.BI \-\-mask2rf " <f>"
+Read in the 'mask' file 
+.I <f>
+and use it to define new #=GC RF annotation for the 
+alignment.
+.I <f>
+must be a single line, with exactly 
+.I <alen> 
+or 
+.I <rflen>
+characters, either the full alignment length or the number of nongap #=GC RF characters, respectively.
+Each character must be either a '1'
+or a '0'. The new #=GC RF markup will contain an 'x' for each column
+that is a '1' in lane mask file, and a '.' for each column that is a '0'. 
+If the mask is of length
+.I <rflen>
+then it is interpreted as applying to only nongap RF characters in the
+existing RF annotation, all gap RF characters will remain gaps and
+nongap RF characters will be redefined as above.
+
+.TP 
+.BI \-\-m\-keeprf
+With 
+.B \-\-mask2rf,
+do not overwrite existing nongap RF characters that are included by
+the input mask as 'x', leave them as the character they are.
+
+.TP 
+.BI \-\-num\-all 
+Add annotation to the alignment numbering all of the columns in the
+alignment. 
+
+.TP 
+.BI \-\-num\-rf 
+Add annotation to the alignment numbering the non-gap (non '.') #=GC
+RF columns of the alignment. 
+
+.TP 
+.BI \-\-rm\-gc " <s>"
+Remove certain types of #=GC annotation from the alignment. 
+.I "<s>" 
+must be one of:
+.BR RF ,
+.BR SS_cons ,
+.BR SA_cons ,
+.BR PP_cons .
+
+.TP 
+.BI \-\-sindi 
+Annotate individual secondary structures for each sequence by imposing
+the consensus secondary structure defined by the #=GC SS_cons
+annotation. 
+
+.TP 
+.BI \-\-post2pp 
+Update Infernal's cmalign 0.72-1.0.2 posterior probability "POST"
+annotation to "PP" annotation, which is read by other miniapps,
+including 
+.B esl\-alimask
+and 
+.B esl\-alistat.
+
+.TP
+.B \-\-amino
+Assert that the 
+.I msafile 
+contains protein sequences. 
+
+.TP 
+.B \-\-dna
+Assert that the 
+.I msafile 
+contains DNA sequences. 
+
+.TP 
+.B \-\-rna
+Assert that the 
+.I msafile 
+contains RNA sequences. 
+
+
+
+.SH SEE ALSO
+
+.nf
+http://bioeasel.org/
+.fi
+
+.SH COPYRIGHT
+
+.nf 
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi 
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi

flax_model/alphafold3/_tools/hmmer/share/man/man1/esl-alimap.1

@@ -0,0 +1,211 @@
+.TH "esl\-alimap" 1 "Aug 2023" "Easel 0.49" "Easel Manual"
+
+.SH NAME
+esl\-alimap \- map two alignments to each other
+
+.SH SYNOPSIS
+.B esl\-alimap
+[\fIoptions\fR]
+.I msafile1
+.I msafile2
+
+.SH DESCRIPTION
+
+.B esl\-alimap
+is a highly specialized application that determines the optimal
+alignment mapping of columns between two alignments of the same
+sequences. An alignment mapping defines for each column in alignment 1
+a matching column in alignment 2. The number of residues in the
+aligned sequences that are in common between the two matched columns
+are considered 'shared' by those two columns.
+
+.PP
+For example, if the nth residue of sequence i occurs in alignment 1
+column x and alignment 2 column y, then only a mapping of alignment
+1 and 2 that includes column x mapping to column y would correctly map
+and share the residue. 
+
+.PP
+The optimal mapping of the two alignments is the mapping which
+maximizes the sum of shared residues between all pairs of matching
+columns. The fraction of total residues that are shared is reported as
+the coverage in the 
+.B esl\-alimap
+output.
+
+.PP
+Only the first alignments in 
+.I msafile1 
+and
+.I msafile2
+will be mapped to each other. If the files contain more than one
+alignment, all alignments after the first will be ignored.
+
+.PP
+The two alignments (one from each file) must contain exactly the same
+sequences (if they were unaligned, they'd be identical) in precisely
+the same order. They must also be in Stockholm format.
+
+.PP
+The output of 
+.B esl\-alimap
+differs depending on whether one or both of the alignments 
+contain reference (#=GC RF) annotation. If so, the
+coverage for residues from nongap RF positions will be reported
+separately from the total coverage.
+
+.PP
+.B esl\-alimap
+uses a dynamic programming algorithm to compute the optimal
+mapping. The algorithm is similar to the Needleman-Wunsch-Sellers
+algorithm but the scores used at each step of the recursion are not
+residue-residue comparison scores but rather the number of residues
+shared between two columns. 
+
+The
+.BI \-\-mask\-a2a " <f>",
+.BI \-\-mask\-a2rf " <f>",
+.BI \-\-mask\-rf2a " <f>",
+and
+.BI \-\-mask\-rf2rf " <f>"
+options create 'mask' files that pertain to the optimal mapping in
+slightly different ways. A mask file consists of a single line, of
+only '0' and '1' characters. These denote which positions of the
+alignment from 
+.I msafile1
+map to positions of the alignment from 
+.I msafile2
+as described below for each of the four respective masking options.
+These masks can be used to extract only those columns of the 
+.I msafile1
+alignment 
+that optimally map to columns of the 
+.I msafile2
+alignment
+using the 
+.B esl\-alimask
+miniapp. To extract the corresponding set of columns 
+from 
+.I msafile2
+(that optimally map to columns of the alignment from
+.IR msafile1 ),
+it is necessary to rerun the program with the order of the two 
+msafiles reversed, save new masks, and use
+.B esl\-alimask
+again.
+
+.SH OPTIONS
+
+.TP
+.B \-h
+Print brief help; includes version number and summary of
+all options.
+
+.TP
+.B \-q
+Be quiet; don't print information the optimal mapping of each column,
+only report coverage and potentially save masks to optional output files. 
+
+.TP
+.BI \-\-mask\-a2a " <f>"
+Save a mask of '0's and '1's to file
+.I <f>.
+A '1' at position x means that position x of the alignment from
+.I msafile1
+maps to an alignment position in the alignment from
+.I msafile2
+in the optimal map.
+
+.TP
+.BI \-\-mask\-a2rf " <f>"
+Save a mask of '0's and '1's to file
+.I <f>.
+A '1' at position x means that position x of the alignment from
+.I msafile1
+maps to a nongap RF position in the alignment from 
+.I msafile2
+in the optimal map.
+
+.TP
+.BI \-\-mask\-rf2a " <f>"
+Save a mask of '0's and '1's to file
+.I <f>.
+A '1' at position x means that nongap RF position x of the alignment from
+.I msafile1
+maps to an alignment position in the alignment from 
+.I msafile2
+in the optimal map.
+
+.TP
+.BI \-\-mask\-rf2rf " <f>"
+Save a mask of '0's and '1's to file
+.I <f>.
+A '1' at position x means that nongap RF position x of the alignment from
+.I msafile1
+maps to a nongap RF position in the alignment from 
+.I msafile2
+in the optimal map.
+
+.TP
+.BI \-\-submap " <f>"
+Specify that all of the columns from the alignment from 
+.I msafile1
+exist identically (contain the same residues from all sequences) in
+the alignment from 
+.I msafile2. 
+This makes the task of mapping trivial.
+However, not all columns of 
+.I msafile1 
+must exist in 
+.I msafile2.
+Save the mask to file
+.I <f>.
+A '1' at position x of the mask means that position x of the alignment from
+.I msafile1
+is the same as position y of
+.I msafile2,
+where y is the number of '1's that occur at positions <= x in the mask.
+
+.TP
+.B \-\-amino
+Assert that 
+.I msafile1
+and 
+.I msafile2
+contain protein sequences. 
+
+.TP 
+.B \-\-dna
+Assert that 
+.I msafile1
+and 
+.I msafile2
+contain DNA sequences. 
+
+.TP 
+.B \-\-rna
+Assert that the 
+.I msafile1
+and 
+.I msafile2
+contain RNA sequences. 
+
+
+.SH SEE ALSO
+
+.nf
+http://bioeasel.org/
+.fi
+
+.SH COPYRIGHT
+
+.nf 
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi 
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi

flax_model/alphafold3/_tools/hmmer/share/man/man1/esl-alimerge.1

@@ -0,0 +1,184 @@
+.TH "esl\-alimerge" 1 "Aug 2023" "Easel 0.49" "Easel Manual"
+
+.SH NAME
+esl\-alimerge \- merge alignments based on their reference (RF) annotation
+
+.SH SYNOPSIS
+
+.nf
+\fBesl\-alimerge \fR[\fIoptions\fR] \fIalifile1 alifile2\fR
+  (merge two alignment files)
+
+\fBesl\-alimerge \-\-list \fR[\fIoptions\fR] \fIlistfile\fR
+  (merge many alignment files listed in a file)
+
+
+.SH DESCRIPTION
+
+.PP
+.B esl\-alimerge
+reads more than one input alignments, merges them into a single
+alignment and outputs it.
+
+.PP
+The input alignments must all be in Stockholm format.  All alignments
+must have reference ('#=GC RF') annotation. Further, the RF annotation
+must be identical in all alignments once gap characters in the RF
+annotation ('.','\-','_') have been removed.  This requirement allows
+alignments with different numbers of total columns to be merged
+together based on consistent RF annotation, such as alignments created
+by successive runs of the
+.B cmalign
+program of the INFERNAL package using the same CM.  Columns which have
+a gap character in the RF annotation are called 'insert' columns.
+
+.PP
+All sequence data in all input alignments will be included in the
+output alignment regardless of the output format (see
+.B \-\-outformat 
+option below). However, sequences in the merged alignment will usually
+contain more gaps ('.') than they did in their respective input
+alignments. This is because 
+.B esl\-alimerge
+must add 100% gap columns to each individual input alignment so that
+insert columns in the other input alignments can be accomodated in the
+merged alignment.
+
+.PP
+If the output format is Stockholm or Pfam, annotation will be
+transferred from the input alignments to the merged alignment as
+follows. All per-sequence ('#=GS') and per-residue ('#=GR') annotation
+is transferred.  Per-file ('#=GF') annotation is transferred if it is
+present and identical in all alignments.  Per-column ('#=GC') annotation is
+transferred if it is present and identical in all alignments once all
+insert positions have been removed and 
+the '#=GC' annotation includes zero non-gap characters in insert
+columns.
+
+.PP
+With the 
+.BI \-\-list " <f>"
+option, 
+.I <f>
+is a file listing alignment files to merge. In the list file, blank
+lines and lines that start with '#' (comments) are ignored. Each data
+line contains a single word: the name of an alignment file to be
+merged. All alignments in each file will be merged.
+
+.PP
+With the
+.B \-\-small
+option, 
+.B esl\-alimerge
+will operate in memory saving mode and the required RAM for the merge
+will be minimal (should be only a few Mb) and independent of the
+alignment sizes. To use 
+.BR \-\-small ,
+all alignments must be in Pfam format (non-interleaved, 1
+line/sequence Stockholm format). You can reformat alignments to Pfam
+using the
+.B esl\-reformat
+Easel miniapp. Without 
+.B \-\-small
+the required RAM will be equal to roughly the size of the final merged
+alignment file which will necessarily be at least the summed size of
+all of the input alignment files to be merged and sometimes several
+times larger. If you're merging large alignments or you're
+experiencing very slow performance of
+.BR esl\-alimerge ,
+try reformatting to Pfam and using
+.BR \-\-small .
+
+
+
+.SH OPTIONS
+
+.TP
+.B \-h
+Print brief help; includes version number and summary of
+all options, including expert options.
+
+.TP
+.BI \-o " <f>"
+Output merged alignment to file 
+.I <f>
+instead of to stdout.
+
+.TP
+.B \-v
+Be verbose; print information on the size of the alignments being merged,
+and the annotation transferred to the merged alignment to stdout.
+This option can only be used in combination with the
+.B \-o 
+option (so that the printed info doesn't corrupt the output alignment
+file).
+
+.TP
+.B \-\-small
+Operate in memory saving mode. Required RAM will be independent of the
+sizes of the alignments to merge, instead of roughly the size of the
+eventual merged alignment. When enabled, all alignments must be in
+Pfam Stockholm (non-interleaved 1 line/seq) format; see
+.BR esl\-reformat (1).
+The output alignment will be in Pfam format.
+
+.TP
+.B \-\-rfonly
+Only include columns that are not gaps in the GC RF annotation in the
+merged alignment. 
+
+.TP 
+.BI \-\-outformat " <s>"
+Write the output alignment in format
+.IR <s> .
+Common choices for 
+.I <s> 
+include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+The string
+.I <s>
+is case-insensitive (\fBa2m\fR or \fBA2M\fR both work).
+Default is
+.BR stockholm .
+
+
+.TP 
+.B \-\-rna
+Specify that the input alignments are RNA alignments. By default
+.B esl\-alimerge
+will try to autodetect the alphabet, but if the alignment is sufficiently
+small it may be ambiguous. This option defines the alphabet as RNA.
+
+.TP 
+.B \-\-dna
+Specify that the input alignments are DNA alignments.
+
+.TP 
+.B \-\-amino
+Specify that the input alignments are protein alignments.
+
+
+
+.SH SEE ALSO
+
+.nf
+http://bioeasel.org/
+.fi
+
+.SH COPYRIGHT
+
+.nf 
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi 
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi

flax_model/alphafold3/_tools/hmmer/share/man/man1/esl-alipid.1

@@ -0,0 +1,137 @@
+.TH "esl\-alipid" 1  "Aug 2023" "Easel 0.49" "Easel Manual"
+
+.SH NAME
+esl\-alipid \- calculate pairwise percent identities for all sequence pairs in an MSA
+
+.SH SYNOPSIS
+.B esl\-alipid
+[\fIoptions\fR]
+.I msafile
+
+
+.SH DESCRIPTION
+
+.PP
+.B esl\-alistat 
+calculates the pairwise percent identity of each sequence pair in
+in the MSA(s) in 
+.I msafile.
+For each sequence pair, it outputs a line of 
+.I <seqname1> <seqname2> <%id> <nid> <denomid> <%match> <nmatch> <denommatch>
+where 
+.I <%id> 
+is the percent identity,
+.I <nid>
+is the number of identical aligned pairs,
+and 
+.I <denomid> 
+is the denominator used for the calculation: the
+shorter of the two (unaligned) sequence lengths.
+The %identity is defined as 100*nid/denomid.
+
+.PP
+The last three fields are the pairwise percent match calculation, in
+the pair\-HMM sense of a "match state" that aligns two residues XY
+(whether identical or different) versus delete \-Y and insert X\- states
+that have a residue in one sequence and a gap character in the other.
+That is, the %match is the percentage of the alignment that
+consists of aligned residues as opposed to insertions or deletions in
+either sequence. The %match is defined as 100*nmatch/denommatch.
+
+.PP
+There are many ways that one could choose a denominator for these
+percentages. We always define %id using MIN(len1,len2) as the
+denominator. In multiple sequence alignments, you will often have short
+sequence fragments which may have very little overlap, or even none at
+all. Several ways to calculate %identity, such as ignoring columns
+with gaps (100* n_identities / (n_identities + n_mismatches)) , or
+dividing by the total alignment length (100 * n_identities / ali_len),
+are not robust to having overlapping fragments or long indels, because
+you can get spuriously high or low %id's.
+
+.PP
+For both %identity and %match calculations, alignments of a gap
+character in both sequences, \-\-,  aren't counted. Also, if the denominator
+is zero (which can happen, when two sequence fragments in the same MSA
+don't overlap each other), the resulting % is defined to be 0.
+
+.PP
+If
+.I msafile 
+is \- (a single dash), alignment input is read from 
+stdin.
+
+.PP
+Only canonical residues are counted toward
+.I <nid> 
+and 
+.I <n>.
+Degenerate residue codes are not counted.
+
+.SH OPTIONS
+
+.TP
+.B \-h 
+Print brief help;  includes version number and summary of
+all options, including expert options.
+
+.TP
+.BI \-\-informat " <s>"
+Assert that input
+.I msafile
+is in alignment format
+.IR <s> ,
+bypassing format autodetection.
+Common choices for 
+.I <s> 
+include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+For more information, and for codes for some less common formats,
+see main documentation.
+The string
+.I <s>
+is case-insensitive (\fBa2m\fR or \fBA2M\fR both work).
+
+.TP
+.B \-\-amino
+Assert that the 
+.I msafile 
+contains protein sequences. 
+
+.TP 
+.B \-\-dna
+Assert that the 
+.I msafile 
+contains DNA sequences. 
+
+.TP 
+.B \-\-rna
+Assert that the 
+.I msafile 
+contains RNA sequences. 
+
+
+
+.SH SEE ALSO
+
+.nf
+http://bioeasel.org/
+.fi
+
+.SH COPYRIGHT
+
+.nf 
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi 
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi

flax_model/alphafold3/_tools/hmmer/share/man/man1/esl-alirev.1

@@ -0,0 +1,153 @@
+.TH "esl\-alirev" 1 "Aug 2023" "Easel 0.49" "Easel Manual"
+
+.SH NAME
+esl\-alirev \- reverse complement a multiple alignment 
+
+
+.SH SYNOPSIS
+.B esl\-alirev
+[\fIoptions\fR]
+.I msafile
+
+
+.SH DESCRIPTION
+
+.PP
+.B esl\-alirev
+reads the multiple alignment in
+.I msafile
+and outputs its reverse complement to stdout.
+
+.PP
+An example of where you might need to do this is when you've
+downloaded a chunk of multiway genomic alignment from one of the
+genome browsers, but your RNA of interest is on the opposite strand.
+
+.PP
+Any per-column and per-residue annotation lines are reversed as well,
+including Stockholm format and old SELEX format annotations.
+Annotations that Easel recognizes as secondary structure annotation (a
+consensus structure line, individual secondary structure lines) will
+be "reverse complemented" to preserve proper bracketing orders: for
+example, ...<<<...>>> is reverse complemented to <<<...>>>..., not
+simply reversed to >>>...<<<..., which would be wrong.
+
+.PP
+If
+.I msafile 
+is \- (a single dash), alignment input is read from stdin.
+
+
+.PP
+By default the output alignment is written in the same format as the
+input alignment. See the
+.B \-\-outformat
+option to use a different output format.
+
+.PP
+Because the alignment is parsed into Easel's digital internal
+representation, the output alignment may differ in certain details
+from the original alignment; these details should be inconsequential
+but may catch your eye. One is that if you have a reference annotation
+line, Easel's output will put consensus residues in upper case,
+nonconsensus (inserted) residues in lower case. Another is that the
+headers for some formats, such as Clustal format, are written with an
+arbitrary version number - so you may find yourself revcomping an
+alignment in "MUSCLE (3.7) multiple sequence alignment" format and it
+could come out claiming to be a "CLUSTAL 2.1 multiple sequence
+alignment", just because Easel writes all of its Clustal format
+alignment files with that header.
+
+.PP
+The
+.I msafile
+must contain nucleic acid sequences (DNA or RNA).
+The alphabet will be autodetected by default.
+See the
+.B \-\-dna
+or
+.B \-\-rna 
+options to assert an alphabet.
+
+
+
+.SH OPTIONS
+
+.TP
+.B \-h 
+Print brief help;  includes version number and summary of
+all options, including expert options.
+
+.TP
+.BI \-\-informat " <s>"
+Assert that input
+.I msafile
+is in alignment format
+.IR <s> ,
+bypassing format autodetection.
+Common choices for 
+.I <s> 
+include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+For more information, and for codes for some less common formats,
+see main documentation.
+The string
+.I <s>
+is case-insensitive (\fBa2m\fR or \fBA2M\fR both work).
+
+.TP
+.BI \-\-outformat " <s>"
+Write the output alignment in alignment format
+.IR <s> .
+Common choices for 
+.I <s> 
+include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+The string
+.I <s>
+is case-insensitive (\fBa2m\fR or \fBA2M\fR both work).
+Default is to use same format as the input
+.IR msafile .
+
+.TP 
+.B \-\-dna
+Assert that the 
+.I msafile 
+contains DNA sequences. 
+
+.TP 
+.B \-\-rna
+Assert that the 
+.I msafile 
+contains RNA sequences. 
+
+
+.SH SEE ALSO
+
+.nf
+http://bioeasel.org/
+.fi
+
+.SH COPYRIGHT
+
+.nf 
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi 
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi
+

flax_model/alphafold3/_tools/hmmer/share/man/man1/esl-alistat.1

@@ -0,0 +1,259 @@
+.TH "esl\-alistat" 1  "Aug 2023" "Easel 0.49" "Easel Manual"
+
+.SH NAME
+esl\-alistat \- summarize a multiple sequence alignment file
+
+.SH SYNOPSIS
+.B esl\-alistat
+[\fIoptions\fR]
+.I msafile
+
+.SH DESCRIPTION
+
+.PP
+.B esl\-alistat 
+summarizes the contents of the multiple sequence alignment(s) in 
+.I msafile, 
+such as the alignment name, format, alignment length (number of
+aligned columns), number of sequences, average pairwise % identity,
+and mean, smallest, and largest raw (unaligned) lengths of the
+sequences.
+
+.PP
+If 
+.I msafile
+is \- (a single dash),
+multiple alignment input is read from stdin.
+
+
+
+.PP
+The 
+.BR \-\-list ,
+.BR \-\-icinfo ,
+.BR \-\-rinfo ,
+.BR \-\-pcinfo ,
+.BR \-\-psinfo ,
+.BR \-\-cinfo ,
+.BR \-\-bpinfo ,
+and
+.B \-\-iinfo
+options allow dumping various statistics on the alignment to optional
+output files as described for each of those options below.
+
+.PP
+The 
+.B \-\-small
+option allows summarizing alignments without storing them in memory
+and can be useful for large alignment files with sizes that approach
+or exceed the amount of available RAM.  When
+.B \-\-small
+is used, 
+.B esl\-alistat
+will print fewer statistics on the alignment, omitting data on the
+smallest and largest sequences and the average identity of the
+alignment.
+.B \-\-small
+only works on Pfam formatted alignments (a special type of
+non-interleaved Stockholm alignment in which each sequence occurs on a
+single line) and 
+.B \-\-informat pfam
+must be given with
+.BR \-\-small .
+Further, when 
+.B \-\-small
+is used, the alphabet must be specified with
+.BR \-\-amino ,
+.BR \-\-dna ,
+or 
+.BR \-\-rna .
+
+
+
+.SH OPTIONS
+
+.TP
+.B \-h 
+Print brief help;  includes version number and summary of
+all options, including expert options.
+
+.TP 
+.B \-1
+Use a tabular output format with one line of statistics per alignment
+in 
+.I msafile.
+This is most useful when
+.I msafile
+contains many different alignments (such as a Pfam database in
+Stockholm format).
+
+
+.SH EXPERT OPTIONS
+
+.TP
+.BI \-\-informat " <s>"
+Assert that input
+.I msafile
+is in alignment format
+.IR <s> ,
+bypassing format autodetection.
+Common choices for 
+.I <s> 
+include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+For more information, and for codes for some less common formats,
+see main documentation.
+The string
+.I <s>
+is case-insensitive (\fBa2m\fR or \fBA2M\fR both work).
+
+
+.TP
+.B \-\-amino
+Assert that the 
+.I msafile 
+contains protein sequences. 
+
+.TP 
+.B \-\-dna
+Assert that the 
+.I msafile 
+contains DNA sequences. 
+
+.TP 
+.B \-\-rna
+Assert that the 
+.I msafile 
+contains RNA sequences. 
+
+.TP 
+.B \-\-small
+Operate in small memory mode for Pfam formatted alignments.
+.B \-\-informat pfam
+and one of
+.BR \-\-amino ,
+.BR \-\-dna ,
+or
+.B \-\-rna
+must be given as well.
+
+.TP 
+.BI \-\-list " <f>"
+List the names of all sequences in all alignments in 
+.B msafile
+to file
+.IR <f> .
+Each sequence name is written on its own line. 
+
+.TP 
+.BI \-\-icinfo " <f>"
+Dump the information content per position in tabular format to file
+.IR <f> .
+Lines prefixed with "#" are comment lines, which explain the
+meanings of each of the tab-delimited fields.
+
+.TP 
+.BI \-\-rinfo " <f>"
+Dump information on the frequency of gaps versus nongap residues per position in tabular format to file
+.IR <f> .
+Lines prefixed with "#" are comment lines, which explain the
+meanings of each of the tab-delimited fields.
+
+.TP 
+.BI \-\-pcinfo " <f>"
+Dump per column information on posterior probabilities in tabular format to file
+.IR <f> .
+Lines prefixed with "#" are comment lines, which explain the
+meanings of each of the tab-delimited fields.
+
+.TP 
+.BI \-\-psinfo " <f>"
+Dump per sequence information on posterior probabilities in tabular format to file
+.IR <f> .
+Lines prefixed with "#" are comment lines, which explain the
+meanings of each of the tab-delimited fields.
+
+.TP 
+.BI \-\-iinfo " <f>"
+Dump information on inserted residues in tabular format to file
+.IR <f> .
+Insert columns of the alignment are those that are gaps in the
+reference (#=GC RF) annotation. This option only works if the input
+file is in Stockholm format with reference annotation.
+Lines prefixed with "#" are comment lines, which explain the
+meanings of each of the tab-delimited fields. 
+
+.TP 
+.BI \-\-cinfo " <f>"
+Dump per-column residue counts to file
+.IR <f> .
+If used in combination with
+.B \-\-noambig
+ambiguous (degenerate) residues will be ignored and not
+counted. Otherwise, they will be marginalized. For example, in an RNA
+sequence file, a 'N' will be counted as 0.25 'A', 0.25 'C', 0.25 'G',
+and 0.25 'U'.
+
+.TP 
+.B \-\-noambig
+With 
+.BR \-\-cinfo ,
+do not count ambiguous (degenerate) residues. 
+
+.TP 
+.B \-\-bpinfo
+Dump per-column basepair counts to file
+.IR <f> .
+Counts appear for each basepair in the consensus secondary structure (annotated as
+"#=GC SS_cons"). Only basepairs from sequences for which both paired positions are
+canonical residues will be counted. That is, any basepair that is a gap
+or an ambiguous (degenerate) residue at either position of the pair is
+ignored and not counted.
+
+
+.TP 
+.B \-\-weight
+With 
+.BR \-\-icinfo ,
+.BR \-\-rinfo ,
+.BR \-\-pcinfo ,
+.BR \-\-iinfo ,
+.BR \-\-cinfo ,
+and
+.BR \-\-bpinfo ,
+weight counts based on #=GS WT annotation in the input 
+.IR msafile .
+A residue or basepair from a sequence with a weight of 
+.I <x>
+will be considered 
+.I <x>
+counts. 
+By default, raw, unweighted counts are reported; corresponding to each
+sequence having an equal weight of 1.
+
+
+
+
+.SH SEE ALSO
+
+.nf
+http://bioeasel.org/
+.fi
+
+.SH COPYRIGHT
+
+.nf 
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi 
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi

flax_model/alphafold3/_tools/hmmer/share/man/man1/esl-compalign.1

@@ -0,0 +1,166 @@
+.TH "esl\-compalign" 1 "Aug 2023" "Easel 0.49" "Easel Manual"
+
+.SH NAME
+esl\-compalign \- compare two multiple sequence alignments
+
+.SH SYNOPSIS
+.B esl\-compalign
+[\fIoptions\fR]
+.I trusted_file
+.I test_file
+
+
+
+.SH DESCRIPTION
+
+.PP
+.B esl\-compalign
+evaluates the accuracy of a predicted multiple sequence alignment with
+respect to a trusted alignment of the same sequences. 
+
+.PP
+The 
+.I trusted_file 
+and
+.I test_file
+must contain the same number of alignments. Each predicted alignment in the 
+.I test_file 
+will be compared against a single trusted alignment from the
+.I trusted_file.
+The first alignments in each file correspond to each other and will be
+compared, the second alignment in each file correspond to each other
+and will be compared, and so on.  Each corresponding pair of
+alignments must contain the same sequences (i.e. if they were
+unaligned they would be identical) in the same order in both
+files. Further, both alignment files must be in Stockholm format and
+contain 'reference' annotation, which appears as "#=GC RF" per-column
+markup for each alignment. The number of nongap (non '.' characters)
+in the reference (RF) annotation must be identical between all
+corresponding alignments in the two files.
+
+.PP
+.B esl\-compalign
+reads an alignment from each file, and compares them based on
+their 'reference' annotation.  The number of correctly predicted
+residues for each sequence is computed as follows. A residue that is
+in the Nth nongap RF column in the trusted alignment must also appear
+in the Nth nongap RF column in the predicted alignment to be counted
+as 'correct', otherwise it is 'incorrect'. A residue that appears in a
+gap RF column in the trusted alignment between nongap RF columns N and
+N+1 must also appear in a nongap RF column in the predicted alignment
+between nongap RF columns N and N+1 to be counted as 'correct',
+otherwise it is incorrect.
+
+.PP
+The default output of
+.B esl\-compalign
+lists each sequence and the number of correctly and incorrectly
+predicted residues for that sequence. These counts are broken down
+into counts for residues in the predicted alignments that occur
+in 'match' columns and 'insert' columns. A 'match' column is one for
+which the RF annotation does not contain a gap. An 'insert' column is
+one for which the RF annotation does contain a gap.
+
+
+
+.SH OPTIONS
+
+.TP
+.B \-h
+Print brief help; includes version number and summary of
+all options.
+
+.TP
+.B \-c
+Print per-column statistics instead of per-sequence statistics.
+
+.TP
+.B \-p 
+Print statistics on accuracy versus posterior probability values. The 
+.I test_file
+must be annotated with posterior probabilities (#=GR PP) for this
+option to work.
+
+
+.SH EXPERT OPTIONS
+
+.TP
+.BI \-\-p\-mask " <f>"
+This option may only be used in combination with the 
+.B \-p
+option. Read a "mask" from file 
+.IR <f> .
+The mask file must consist of a single line, of only '0' and '1'
+characters. There must be exactly RFLEN characters where RFLEN is the
+number of nongap characters in the RF annotation of all alignments in
+both 
+.I trusted_file
+and
+.IR test_file .
+Positions of the mask that are '1' characters indicate that the
+corresponding nongap RF position is included by the mask. The
+posterior probability accuracy statistics for match columns will only
+pertain to positions that are included by the mask, those that are
+excluded will be ignored from the accuracy calculation.
+
+.BI \-\-c2dfile " <f>"
+Save a 'draw file' to file 
+.I <f>
+which can be read into the 
+.B esl\-ssdraw
+miniapp. This draw file will define two postscript pages for 
+.BR esl\-ssdraw .
+The first page will depict the frequency of errors per match position and
+frequency of gaps per match position, indicated by magenta and yellow,
+respectively. The darker magenta, the more errors and the darker
+yellow, the more gaps. The second page will depict the frequency of
+errors in insert positions in shades of magenta, the darker the
+magenta the more errors in inserts after each position. See
+.B esl\-ssdraw
+documentation for more information on these diagrams. 
+
+.TP
+.B \-\-amino
+Assert that 
+.I trusted_file
+and 
+.I test_file
+contain protein sequences. 
+
+.TP 
+.B \-\-dna
+Assert that 
+.I trusted_file
+and 
+.I test_file
+contain DNA sequences. 
+
+.TP 
+.B \-\-rna
+Assert that the 
+.I trusted_file
+and 
+.I test_file
+contain RNA sequences. 
+
+
+
+.SH SEE ALSO
+
+.nf
+http://bioeasel.org/
+.fi
+
+.SH COPYRIGHT
+
+.nf 
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi 
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi
+

flax_model/alphafold3/_tools/hmmer/share/man/man1/esl-histplot.1

@@ -0,0 +1,87 @@
+.TH "esl\-histplot" 1 "Aug 2023" "Easel 0.49" "Easel Manual"
+
+.SH NAME
+esl\-histplot \- collate data histogram, output xmgrace datafile
+
+.SH SYNOPSIS
+.B esl\-histplot
+[\fIoptions\fR]
+.I datafile
+
+
+.SH DESCRIPTION
+
+.PP
+.B esl\-histplot
+summarizes numerical data in the input file
+.I datafile.
+
+.PP
+One real-numbered value is taken from each line of the input file. 
+Each line is split into whitespace-delimited fields, and one field
+is converted to data. By default this is the first field; this can be 
+changed by the 
+.B \-f 
+option.
+
+.PP
+Default output is a survival plot (Prob(value > x)) in xmgrace XY data
+format, to stdout.
+Output may be directed to a file with the
+.B \-o
+option.
+
+.PP
+If 
+.I datafile
+is \- (a single dash), input lines are read from stdin
+instead of opening a file.
+
+
+
+
+.SH OPTIONS
+
+.TP
+.BI \-f  " <n>"
+Read data from whitespace-delimited field 
+.I <n>
+on each line, instead of the first field.
+Fields are numbered starting from 1.
+
+.TP
+.B \-h 
+Print brief help; includes version number and summary of
+all options, including expert options.
+
+.TP
+.BI \-o  " <f>"
+Send output to file
+.I <f>
+instead of stdout.
+
+
+
+.SH SEE ALSO
+
+.nf
+http://bioeasel.org/
+.fi
+
+.SH COPYRIGHT
+
+.nf 
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi 
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi
+
+
+
+
+

flax_model/alphafold3/_tools/hmmer/share/man/man1/esl-mask.1

@@ -0,0 +1,198 @@
+.TH "esl\-mask" 1 "Aug 2023" "Easel @EASELVERSION@" "Easel Manual"
+
+.SH NAME
+esl\-mask \- mask sequence residues with X's (or other characters)
+
+.SH SYNOPSIS
+.B esl\-mask
+[\fIoptions\fR]
+.I seqfile
+.I maskfile
+
+
+.SH DESCRIPTION
+
+.PP
+.B esl\-mask
+reads lines from 
+.I maskfile 
+that give start/end coordinates for
+regions in each sequence in 
+.IR seqfile ,
+masks these residues (changes
+them to X's), and outputs the masked sequence.
+
+.PP
+The
+.I maskfile
+is a space-delimited file. Blank lines and lines that start with '#'
+(comments) are ignored. Each data line contains at least three
+fields:
+.IR seqname ,
+.IR start ,
+and
+.IR end . 
+The 
+.I seqname 
+is the name of a sequence in the 
+.IR seqfile ,
+and 
+.I start
+and 
+.I end 
+are coordinates defining a region in that sequence.
+The coordinates are indexed <1..L> with respect to a
+sequence of length <L>.
+
+.PP
+By default, the sequence names must appear in exactly the same order
+and number as the sequences in the
+.IR seqfile.  
+This is easy to enforce, because the format of
+.I maskfile 
+is also legal as a list of names for 
+.BR esl\-sfetch , 
+so you can always fetch a temporary sequence file with 
+.B esl\-sfetch 
+and pipe that to 
+.BR esl\-mask .
+(Alternatively, see the 
+.B \-R
+option for fetching from an SSI-indexed
+.IR seqfile .)
+
+.PP
+The default is to mask the region indicated by 
+\fI<start>\fR..\fI<end>\fR.
+Alternatively, everything but this region can be masked;
+see the
+.B \-r
+reverse masking option.
+
+.PP
+The default is to mask residues by converting them to X's.
+Any other masking character can be chosen (see
+.B \-m
+option), or alternatively, masked residues can be lowercased (see
+.B \-l
+option).
+
+
+
+.SH OPTIONS
+
+.TP
+.B \-h
+Print brief help; includes version number and summary of
+all options, including expert options.
+
+.TP
+.B \-l
+Lowercase; mask by converting masked characters to lower case and
+unmasked characters to upper case.
+
+.TP
+.BI \-m " <c>"
+Mask by converting masked residues to 
+.I <c>
+instead of the default X.
+
+.TP
+.BI \-o  " <f>"
+Send output to file
+.I <f>
+instead of stdout.
+
+.TP
+.B \-r
+Reverse mask; mask everything outside the region
+.I start..end, 
+as opposed to the default of masking that region.
+
+.TP
+.B \-R
+Random access; 
+fetch sequences from 
+.I seqfile
+rather than requiring that sequence names in
+.I maskfile
+and
+.I seqfile
+come in exactly the same order and number.  The
+.I seqfile
+must be SSI indexed (see \fBesl\-sfetch \-\-index\fR.)
+
+.TP
+.BI \-x " <n>"
+Extend all masked regions by up to <n> residues on each side. 
+For normal masking, this means masking
+\fI<start>\fR\-\fI<n>\fR..\fI<end>\fR+\fI<n>\fR. 
+For reverse masking, this means masking
+1..\fI<start>\fR\-1+\fI<n>\fR
+and
+\fI<end>\fR+1\-\fI<n>\fR..L
+in a sequence of length L.
+
+
+.TP
+.BI \-\-informat " <s>"
+Assert that input
+.I seqfile
+is in format
+.IR <s> ,
+bypassing format autodetection.
+Common choices for 
+.I <s> 
+include:
+.BR fasta ,
+.BR embl ,
+.BR genbank.
+Alignment formats also work;
+common choices include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+For more information, and for codes for some less common formats,
+see main documentation.
+The string
+.I <s>
+is case-insensitive (\fBfasta\fR or \fBFASTA\fR both work).
+
+
+
+
+.SH SEE ALSO
+
+.nf
+http://bioeasel.org/
+.fi
+
+.SH COPYRIGHT
+
+.nf 
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi 
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi
+
+
+
+
+
+
+
+
+
+
+
+
+
+

flax_model/alphafold3/_tools/hmmer/share/man/man1/esl-mixdchlet.1

@@ -0,0 +1,212 @@
+.TH "esl\-mixdchlet" 1 "Aug 2023" "Easel 0.49" "Easel Manual"
+
+.SH NAME
+esl\-mixdchlet \- fitting mixture Dirichlets to count data
+
+.SH SYNOPSIS
+
+.nf
+\fBesl\-mixdchlet fit\fR [\fIoptions\fR] \fIQ K in_countfile out_mixchlet\fR
+  (train a new mixture Dirichlet)
+
+\fBesl\-mixdchlet score\fR [\fIoptions\fR] \fImixdchlet_file counts_file\fR
+  (calculate log likelihood of count data, given mixture Dirichlet)
+
+\fBesl\-mixdchlet gen \fR[\fIoptions\fR] \fImixdchlet_file\fR
+  (generate synthetic count data from mixture Dirichlet)
+
+\fBesl\-mixdchlet sample \fR[\fIoptions\fR]
+  (sample a random mixture Dirichlet for testing)
+.fi
+
+
+.SH DESCRIPTION
+
+.PP
+The
+.B esl\-mixdchlet
+miniapp is for training mixture Dirichlet priors, such as the priors
+used in HMMER and Infernal. It has four subcommands:
+.B fit,
+.B score,
+.B gen,
+and
+.B sample.
+The most important subcommand is
+.B fit,
+which is the subcommand for fitting a new mixture Dirichlet
+distribution to a collection of count vectors (for example,
+emission or transition count vectors from Pfam or Rfam training
+sets).
+
+.PP
+Specifically,
+.B esl\-mixdchlet fit
+fits a new mixture Dirichlet distribution with
+.I Q
+mixture components to the count vectors (of alphabet size
+.I K
+) in input file
+.I in_countfile,
+and saves the mixture Dirichlet into output file
+.I out_mixdchlet.
+
+.PP
+The input count vector file
+.I in_countfile
+contains one count vector of length
+.I K
+fields per line, for any number of lines.
+Blank lines and lines starting in # (comments) are ignored.
+Fields are nonnegative real values; they do not have to be integers,
+because they can be weighted counts.
+
+.PP
+The format of a mixture Dirichlet file
+.I out_mixdchlet
+is as follows. The first line has two fields,
+.I K Q,
+where
+.I K
+is the alphabet size and 
+.I Q
+is the number of mixture components.
+The next
+.I Q
+lines consist of
+.I K+1
+fields. The first field is the mixture coefficient
+.I q_k,
+followed by
+.I K
+fields with the Dirichlet alpha[k][a] parameters
+for this component.
+
+.PP
+The
+.B esl\-mixdchlet score
+subcommand calculates the log likelihood of the count vector data in
+.I counts_file,
+given the mixture Dirichlet in
+.I mixdchlet_file.
+
+.PP
+The
+.B esl\-mixdchlet gen
+subcommand generates synthetic count data, given
+a mixture Dirichlet.
+
+.PP
+The
+.B esl\-mixdchlet sample
+subcommand creates a random mixture Dirichlet distribution 
+and outputs it to standard output.
+
+
+.SH OPTIONS FOR FIT SUBCOMMAND
+
+.TP
+.B \-h
+Print brief help specific to the
+.B fit
+subcommand.
+
+.TP
+.BI \-s " <seed>"
+Set random number generator seed to nonnegative integer
+.I <seed>.
+Default is 0, which means to use a quasirandom arbitrary seed.
+Values >0 give reproducible results.
+
+
+
+
+.SH OPTIONS FOR SCORE SUBCOMMAND
+
+.TP
+.B \-h
+Print brief help specific to the
+.B score
+subcommand.
+
+
+
+.SH OPTIONS FOR GEN SUBCOMMAND
+
+.TP
+.B \-h
+Print brief help specific to the
+.B gen
+subcommand.
+
+.TP
+.BI \-s " <seed>"
+Set random number generator seed to nonnegative integer
+.I <seed>.
+Default is 0, which means to use a quasirandom arbitrary seed.
+Values >0 give reproducible results.
+
+
+.TP
+.BI \-M " <M>"
+Generate
+.I <M>
+counts per sampled vector. (Default 100.)
+
+.TP
+.BI \-N " <N>"
+Generate
+.I <N>
+count vectors. (Default 1000.)
+
+
+.SH OPTIONS FOR SAMPLE SUBCOMMAND
+
+.TP
+.B \-h
+Print brief help specific to the
+.B sample
+subcommand.
+
+.TP
+.BI \-s " <seed>"
+Set random number generator seed to nonnegative integer
+.I <seed>.
+Default is 0, which means to use a quasirandom arbitrary seed.
+Values >0 give reproducible results.
+
+
+.TP
+.BI \-K " <K>"
+Set the alphabet size to 
+.I <K>.
+(Default is 20, for amino acids.)
+
+.TP
+.BI \-Q " <Q>"
+Set the number of mixture components to
+.I <Q>.
+(Default is 9.)
+
+
+
+
+.SH SEE ALSO
+
+.nf
+http://bioeasel.org/
+.fi
+
+.SH COPYRIGHT
+
+.nf 
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi 
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi
+

flax_model/alphafold3/_tools/hmmer/share/man/man1/esl-reformat.1

@@ -0,0 +1,271 @@
+.TH "esl\-reformat" 1 "Aug 2023" "Easel 0.49" "Easel Manual"
+
+.SH NAME
+esl\-reformat \- convert sequence file formats
+
+.SH SYNOPSIS
+.B esl\-reformat
+[\fIoptions\fR]
+.I format
+.I seqfile
+
+
+.SH DESCRIPTION
+
+.PP
+.B esl\-reformat
+reads the sequence file
+.I seqfile
+in any supported format, reformats it
+into a new format specified by 
+.IR format ,
+then outputs the reformatted text.
+
+.PP
+The 
+.I format
+argument must (case-insensitively) match a supported sequence file format.
+Common choices for 
+.I format
+include:
+.BR fasta ,
+.BR embl ,
+.BR genbank.
+If
+.I seqfile
+is an alignment file,
+alignment output formats also work.
+Common choices include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+For more information, and for codes for some less common formats,
+see main documentation.
+The string
+.I <s>
+is case-insensitive (\fBfasta\fR or \fBFASTA\fR both work).
+
+.PP
+Unaligned format files cannot be reformatted to
+aligned formats.
+However, aligned formats can be reformatted
+to unaligned formats, in which case gap characters are 
+simply stripped out.
+
+.SH OPTIONS
+
+.TP
+.B \-d 
+DNA; convert U's to T's, to make sure a nucleic acid
+sequence is shown as DNA not RNA. See
+.B \-r.
+
+
+.TP
+.B \-h
+Print brief help; includes version number and summary of
+all options, including expert options.
+
+
+.TP
+.B \-l
+Lowercase; convert all sequence residues to lower case.
+See
+.BR \-u .
+
+
+.TP
+.B \-n
+For DNA/RNA sequences, converts any character that's not unambiguous
+RNA/DNA (e.g. ACGTU/acgtu) to an N. Used to convert IUPAC ambiguity
+codes to N's, for software that can't handle all IUPAC codes (some
+public RNA folding codes, for example). If the file is an alignment,
+gap characters are also left unchanged. If sequences are not
+nucleic acid sequences, this option will corrupt the data in
+a predictable fashion.
+
+
+.TP
+.BI \-o  " <f>"
+Send output to file
+.I <f>
+instead of stdout.
+
+
+.TP
+.B \-r 
+RNA; convert T's to U's, to make sure a nucleic acid
+sequence is shown as RNA not DNA. See
+.BR \-d .
+
+
+.TP
+.B \-u
+Uppercase; convert all sequence residues to upper case.
+See
+.BR \-l .
+
+
+.TP
+.B \-x
+For DNA sequences, convert non-IUPAC characters (such as X's) to N's.
+This is for compatibility with benighted people who insist on using X
+instead of the IUPAC ambiguity character N. (X is for ambiguity
+in an amino acid residue). 
+.IP
+Warning: like the
+.B \-n
+option, the code doesn't check that you are actually giving it DNA. It
+simply literally just converts non-IUPAC DNA symbols to N. So if you
+accidentally give it protein sequence, it will happily convert most
+every amino acid residue to an N.
+
+
+
+
+.SH EXPERT OPTIONS
+
+
+.TP
+.BI \-\-gapsym " <c>"
+Convert all gap characters to 
+.IR <c> .
+Used to prepare alignment files for programs with strict
+requirements for gap symbols. Only makes sense if
+the input 
+.I seqfile
+is an alignment.
+
+.TP
+.BI \-\-informat " <s>"
+Assert that input
+.I seqfile
+is in format
+.IR <s> ,
+bypassing format autodetection.
+Common choices for 
+.I <s> 
+include:
+.BR fasta ,
+.BR embl ,
+.BR genbank.
+Alignment formats also work;
+common choices include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+For more information, and for codes for some less common formats,
+see main documentation.
+The string
+.I <s>
+is case-insensitive (\fBfasta\fR or \fBFASTA\fR both work).
+
+.TP
+.B \-\-mingap
+If 
+.I seqfile
+is an alignment, remove any columns that contain 100% gap or missing
+data characters, minimizing the overall length of the alignment.
+(Often useful if you've extracted a subset of aligned sequences from a
+larger alignment.)
+
+.TP
+.B \-\-keeprf
+When used in combination with
+.BR \-\-mingap ,
+never remove a column that is not a gap in the reference (#=GC RF) 
+annotation, even if the column contains 100% gap characters in 
+all aligned sequences. By default with
+.BR \-\-mingap ,
+nongap RF columns that are 100% gaps in all sequences are removed.
+
+.TP
+.B \-\-nogap
+Remove any aligned columns that contain any gap or missing data
+symbols at all. Useful as a prelude to phylogenetic analyses, where
+you only want to analyze columns containing 100% residues, so you want
+to strip out any columns with gaps in them.  Only makes sense if the
+file is an alignment file.
+
+.TP
+.B \-\-wussify
+Convert RNA secondary structure annotation strings (both consensus
+and individual) from old "KHS" format, ><, to the new WUSS notation,
+<>. If the notation is already in WUSS format, this option will screw it
+up, without warning. Only SELEX and Stockholm format files have
+secondary structure markup at present.
+
+.TP
+.B \-\-dewuss
+Convert RNA secondary structure annotation strings from the new
+WUSS notation, <>, back to the old KHS format, ><. If the annotation
+is already in KHS, this option will corrupt it, without warning.
+Only SELEX and Stockholm format files have secondary structure
+markup.
+
+.TP
+.B \-\-fullwuss
+Convert RNA secondary structure annotation strings from simple
+(input) WUSS notation to full (output) WUSS notation.
+
+.TP 
+.BI \-\-replace " <s>"
+.I <s>
+must be in the format
+.I <s1>:<s2>
+with equal numbers of characters in 
+.I <s1>
+and 
+.I <s2>
+separated by a ":" symbol. Each character from
+.I <s1>
+in the input file will be replaced by its counterpart (at the same
+position) from
+.IR <s2> .
+Note that special characters in 
+.I <s>
+(such as "~") may need to be prefixed by
+a "\\" character. 
+
+.TP
+.B \-\-small
+Operate in small memory mode for input alignment files in 
+Pfam format. If not used, each alignment is stored in memory so the
+required memory will be roughly the size of the largest alignment
+in the input file. With 
+.BR \-\-small , 
+input alignments are not stored in memory. 
+This option only works in combination with 
+.B \-\-informat pfam
+and output format 
+.I pfam
+or
+.IR afa . 
+
+
+
+.SH SEE ALSO
+
+.nf
+http://bioeasel.org/
+.fi
+
+.SH COPYRIGHT
+
+.nf 
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi 
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi
+

flax_model/alphafold3/_tools/hmmer/share/man/man1/esl-selectn.1

@@ -0,0 +1,81 @@
+.TH "esl\-selectn" 1 "Aug 2023" "Easel 0.49" "Easel Manual"
+
+.SH NAME
+esl\-selectn \- select random subset of lines from file
+
+.SH SYNOPSIS
+.B esl\-selectn
+[\fIoptions\fR]
+.I nlines
+.I filename
+
+
+.SH DESCRIPTION
+
+.PP
+.B esl\-selectn
+selects 
+.I nlines
+lines at random from file 
+.I filename
+and outputs them on 
+.I stdout.
+
+.PP
+If 
+.I filename
+is \- (a single dash),
+input is read from stdin.
+
+
+.PP
+Uses an efficient reservoir sampling algorithm that only requires only a single pass through
+.I filename,
+and memory storage proportional to 
+.I nlines
+(and importantly, not to the size of the file
+.I filename
+itself).
+.B esl\-selectn 
+can therefore be used to create large scale statistical sampling 
+experiments, especially in combination with other Easel
+miniapplications.
+
+
+.SH OPTIONS
+
+.TP
+.B \-h
+Print brief help; includes version number and summary of
+all options, including expert options.
+
+
+.TP
+.BI \-\-seed " <d>"
+Set the random number seed to
+.I <d>,
+an integer >= 0.
+The default is 0, which means to use a randomly selected seed.
+A seed > 0 results
+in reproducible identical samples from different runs of the same
+command.
+
+
+.SH SEE ALSO
+
+.nf
+http://bioeasel.org/
+.fi
+
+.SH COPYRIGHT
+
+.nf 
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi 
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi

flax_model/alphafold3/_tools/hmmer/share/man/man1/esl-seqstat.1

@@ -0,0 +1,117 @@
+.TH "esl\-seqstat" 1  "Aug 2023" "Easel 0.49" "Easel Manual"
+
+.SH NAME
+esl\-seqstat \- summarize contents of a sequence file
+
+.SH SYNOPSIS
+.B esl\-seqstat
+[\fIoptions\fR]
+.I seqfile
+
+.SH DESCRIPTION
+
+.PP
+.B esl\-seqstat 
+summarizes the contents of the
+.IR seqfile .
+It prints the format, alphabet type, number of sequences, total number
+of residues, and the mean, smallest, and largest sequence length.
+
+.PP
+If 
+.I seqfile
+is \- (a single dash),
+sequence input is read from stdin.
+
+
+
+
+.SH OPTIONS
+
+.TP
+.B \-h 
+Print brief help;  includes version number and summary of
+all options, including expert options.
+
+.TP
+.B \-a
+Additionally show a summary statistic line showing the name, length,
+and description of each individual sequence. Each of these lines is
+prefixed by an = character, in order to allow these lines to be easily
+grepped out of the output.
+
+.TP
+.B \-c
+Additionally print the residue composition of the sequence file.
+
+
+
+.SH EXPERT OPTIONS
+
+.TP
+.BI \-\-informat " <s>"
+Assert that input
+.I seqfile
+is in format
+.IR <s> ,
+bypassing format autodetection.
+Common choices for 
+.I <s> 
+include:
+.BR fasta ,
+.BR embl ,
+.BR genbank.
+Alignment formats also work;
+common choices include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+For more information, and for codes for some less common formats,
+see main documentation.
+The string
+.I <s>
+is case-insensitive (\fBfasta\fR or \fBFASTA\fR both work).
+
+
+.TP
+.B \-\-amino
+Assert that the 
+.I seqfile 
+contains protein sequences. 
+
+.TP 
+.B \-\-dna
+Assert that the 
+.I seqfile 
+contains DNA sequences. 
+
+.TP 
+.B \-\-rna
+Assert that the 
+.I seqfile 
+contains RNA sequences. 
+
+
+
+.SH SEE ALSO
+
+.nf
+http://bioeasel.org/
+.fi
+
+.SH COPYRIGHT
+
+.nf 
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi 
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi
+

flax_model/alphafold3/_tools/hmmer/share/man/man1/esl-sfetch.1

@@ -0,0 +1,278 @@
+.TH "esl\-sfetch" 1 "Aug 2023" "Easel 0.49" "Easel Manual"
+
+.SH NAME
+esl\-sfetch \- retrieve (sub-)sequences from a sequence file
+
+.SH SYNOPSIS
+
+.nf
+\fBesl\-sfetch\fR [\fIoptions\fR] \fIseqfile key\fR
+  (retrieve a single sequence by key)
+
+\fBesl\-sfetch \-c \fR\fIfrom\fR\fB..\fR\fIto \fR[\fIoptions\fR]\fI seqfile key\fR
+  (retrieve a single subsequence by key and coords)
+
+\fBesl\-sfetch \-f \fR[\fIoptions\fR] \fIseqfile keyfile\fR
+  (retrieve multiple sequences using a file of keys)
+
+\fBesl\-sfetch \-Cf \fR[\fIoptions\fR] \fIseqfile subseq\-coord\-file\fR
+  (retrieve multiple subsequences using file of keys and coords)
+
+\fBesl\-sfetch \-\-index\fR\fI msafile\fR
+  (index a sequence file for retrievals)
+.fi
+
+
+.SH DESCRIPTION
+
+.PP
+.B esl\-sfetch
+retrieves one or more sequences or subsequences from
+.IR seqfile .
+
+.PP
+The 
+.I seqfile 
+must be indexed using
+\fBesl\-sfetch \-\-index\fR\fI seqfile\fR.
+This creates an SSI index file
+.IR seqfile .ssi.
+
+.PP
+To retrieve a single complete sequence, do
+\fBesl\-sfetch\fR\fI seqfile key\fR,
+where 
+.I key
+is the name or accession of the desired sequence.
+
+.PP
+To retrieve a single subsequence rather than a complete
+sequence, use the 
+\fB\-c \fR\fIstart\fR..\fIend\fR
+option to provide
+.I start
+and
+.I end
+coordinates. The
+.I start
+and
+.I end
+coordinates are provided as one string, separated
+by any nonnumeric, nonwhitespace character or characters you like;
+see the
+.B \-c
+option below for more details.
+
+.PP
+To retrieve more than one complete sequence at once, you may use the 
+.B \-f
+option, and the second command line argument will specify the
+name of a 
+.I keyfile
+that contains a list of names or accessions, one per line; the first
+whitespace-delimited field on each line of this file is parsed as the
+name/accession.
+
+.PP
+To retrieve more than one subsequence at once, use the
+.B \-C
+option in addition to
+.BR \-f ,
+and now the second argument is parsed as a list of subsequence
+coordinate lines. See the
+.B \-C
+option below for more details, including the format of these lines.
+
+ 
+.PP
+In DNA/RNA files, you may extract (sub-)sequences in reverse complement
+orientation in two different ways: either by providing a 
+.I from
+coordinate that is greater than 
+.IR to , 
+or by providing the 
+.I \-r
+option.
+
+.PP
+When the
+.B \-f 
+option is used to do multiple (sub-)sequence retrieval, the file
+argument may be \- (a single dash), in which case the list of
+names/accessions (or subsequence coordinate lines) is read from
+standard input. However, because a standard input stream can't be SSI indexed,
+(sub-)sequence retrieval from stdin may be slow.
+
+
+.SH OPTIONS
+
+.TP
+.B \-h
+Print brief help; includes version number and summary of
+all options, including expert options.
+
+.TP
+.BI \-c " coords"
+Retrieve a subsequence with start and end coordinates specified by the 
+.I coords
+string. This string consists of start 
+and end coordinates separated
+by any nonnumeric, nonwhitespace character or characters you like;
+for example, 
+\fB\-c 23..100\fR,
+\fB\-c 23/100\fR, or
+\fB\-c 23\-100\fR
+all work. To retrieve a suffix of a subsequence, you
+can omit the 
+.I end
+; for example,
+.B \-c 23:
+would work.
+To specify reverse complement (for DNA/RNA sequence),
+you can specify 
+.I from
+greater than
+.IR to ;
+for example,
+.B \-c 100..23
+retrieves the reverse complement strand from 100 to 23.
+
+.TP
+.B \-f
+Interpret the second argument as a 
+.I keyfile
+instead of as just one
+.I key. 
+The first whitespace-limited field on each line of 
+.I keyfile
+is interpreted as a name or accession to be fetched.
+This option doesn't work with the
+.B \-\-index
+option.  Any other fields on a line after the first one are
+ignored. Blank lines and lines beginning with # are ignored.
+
+.TP
+.BI \-o " <f>"
+Output retrieved sequences to a file 
+.I <f>
+instead of to stdout.
+
+
+.TP
+.BI \-n " <s>"
+Rename the retrieved (sub-)sequence 
+.IR <s> .
+Incompatible with 
+.BR \-f .
+
+.TP
+.B \-r
+Reverse complement the retrieved (sub-)sequence. Only accepted for
+DNA/RNA sequences.
+
+.TP
+.B \-C
+Multiple subsequence retrieval mode, with 
+.B \-f
+option (required). Specifies that the second command line argument
+is to be parsed as a subsequence coordinate file, consisting of
+lines containing four whitespace-delimited fields:
+.IR new_name ,
+.IR from ,
+.IR to ,
+.IR name/accession .
+For each such line, sequence
+.I name/accession
+is found, a subsequence
+\fIfrom\fR..\fIto\fR is extracted,
+and the subsequence is renamed 
+.I new_name 
+before being output. 
+Any other fields after the first four are ignored. Blank lines
+and lines beginning with # are ignored.
+
+
+.TP
+.B \-O
+Output retrieved sequence to a file named
+.IR key .
+This is a convenience for saving some typing:
+instead of 
+.nf
+  \fB% esl\-sfetch \-o SRPA_HUMAN swissprot SRPA_HUMAN
+.fi
+you can just type
+.nf 
+  \fB% esl\-sfetch \-O swissprot SRPA_HUMAN
+.fi
+The
+.B \-O 
+option only works if you're retrieving a
+single alignment; it is incompatible with 
+.BR \-f .
+
+.TP
+.B \-\-index
+Instead of retrieving a
+.I key,
+the special command
+.B esl\-sfetch \-\-index
+.I seqfile
+produces an SSI index of the names and accessions
+of the alignments in
+the 
+.I seqfile.
+Indexing should be done once on the
+.I seqfile
+to prepare it for all future fetches.
+
+
+.SH EXPERT OPTIONS
+
+.TP
+.BI \-\-informat " <s>"
+Assert that 
+.I seqfile
+is in format
+.IR <s> ,
+bypassing format autodetection.
+Common choices for 
+.I <s> 
+include:
+.BR fasta ,
+.BR embl ,
+.BR genbank.
+Alignment formats also work;
+common choices include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+For more information, and for codes for some less common formats,
+see main documentation.
+The string
+.I <s>
+is case-insensitive (\fBfasta\fR or \fBFASTA\fR both work).
+
+
+
+.SH SEE ALSO
+
+.nf
+http://bioeasel.org/
+.fi
+
+.SH COPYRIGHT
+
+.nf 
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi 
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi

flax_model/alphafold3/_tools/hmmer/share/man/man1/esl-shuffle.1

@@ -0,0 +1,234 @@
+.TH "esl\-shuffle" 1  "Aug 2023" "Easel 0.49" "Easel Manual"
+
+.SH NAME
+esl\-shuffle \- shuffling sequences or generating random ones
+
+.SH SYNOPSIS
+
+.nf
+\fBesl\-shuffle \fR[\fIoptions\fR] \fIseqfile\fR
+  (shuffle sequences)
+
+\fBesl\-shuffle \-G \fR[\fIoptions\fR]
+  (generate random sequences)
+
+\fBesl\-shuffle \-A \fR[\fIoptions\fR] \fImsafile\fR
+  (shuffle multiple sequence alignments)
+.fi
+
+.SH DESCRIPTION
+
+.PP
+.B esl\-shuffle
+has three different modes of operation.
+
+.PP
+By default, 
+.B esl\-shuffle
+reads individual sequences from 
+.IR seqfile ,
+shuffles them, and outputs the shuffled sequences.
+By default, shuffling is done by preserving monoresidue
+composition; other options are listed below.
+
+.PP
+With the 
+.B \-G 
+option,
+.B esl\-shuffle
+generates some number of random sequences of some length in
+some alphabet. The
+.B \-N
+option controls the number (default is 1), the
+.B \-L
+option controls the length (default is 0), 
+and the 
+.BR \-\-amino ,
+.BR \-\-dna ,
+and 
+.B \-\-rna
+options control the alphabet.
+
+.PP
+With the 
+.B \-A
+option, 
+.B esl\-shuffle
+reads one or more multiple alignments from
+.I msafile
+shuffles them, and outputs the shuffled alignments.
+By default, the alignment is shuffled columnwise
+(i.e. column order is permuted).
+Other options are listed below.
+
+
+.SH GENERAL OPTIONS
+
+.TP
+.B \-h 
+Print brief help;  includes version number and summary of
+all options, including expert options.
+
+.TP
+.BI \-o " <f>"
+Direct output to a file named
+.I <f>
+rather than to stdout.
+
+.TP
+.BI \-N " <n>"
+Generate 
+.I <n>
+sequences, or
+.I <n> 
+perform independent shuffles per input sequence or alignment.
+
+.TP
+.BI \-L " <n>"
+Generate sequences of length
+.IR <n> ,
+or truncate output shuffled sequences or alignments to a length of
+.IR <n> .
+
+
+
+
+.SH SEQUENCE SHUFFLING OPTIONS
+
+These options only apply in default (sequence shuffling) mode.  They
+are mutually exclusive.
+
+.TP
+.B \-m
+Monoresidue shuffling (the default): preserve monoresidue composition exactly.
+Uses the Fisher/Yates algorithm (aka Knuth's "Algorithm P").
+
+.TP
+.B \-d
+Diresidue shuffling; preserve diresidue composition exactly.  Uses the
+Altschul/Erickson algorithm (Altschul and Erickson, 1986). A more
+efficient algorithm (Kandel and Winkler 1996) is known but has not yet
+been implemented in Easel.
+
+.TP
+.B \-0
+0th order Markov generation: generate a sequence of the same length
+with the same 0th order Markov frequencies. Such a sequence will
+approximately preserve the monoresidue composition of the input.
+
+.TP
+.B \-1
+1st order Markov generation: generate a sequence of the same length
+with the same 1st order Markov frequencies. Such a sequence will 
+approximately preserve the diresidue composition of the input.
+
+.TP
+.B \-r
+Reversal; reverse each input.
+
+.TP
+.BI \-w " <n>"
+Regionally shuffle the input in nonoverlapping windows of size 
+.I <n> 
+residues, preserving exact monoresidue composition in each window.
+ 
+
+
+.SH MULTIPLE ALIGNMENT SHUFFLING OPTIONS
+
+.TP
+.B \-b
+Sample columns with replacement, in order to generate a
+bootstrap-resampled alignment dataset. 
+
+.TP
+.B \-v
+Shuffle residues with each column independently; i.e., permute residue
+order in each column ("vertical" shuffling).
+
+
+.SH SEQUENCE GENERATION OPTIONS
+
+One of these must be selected, if
+.B \-G
+is used.
+
+.TP
+.B \-\-amino
+Generate amino acid sequences.
+
+.TP 
+.B \-\-dna
+Generate DNA sequences.
+
+.TP 
+.B \-\-rna
+Generate RNA sequences.
+
+
+
+.SH EXPERT OPTIONS
+
+.TP
+.BI \-\-informat " <s>"
+Assert that input
+.I seqfile
+is in format
+.IR <s> ,
+bypassing format autodetection.
+Common choices for 
+.I <s> 
+include:
+.BR fasta ,
+.BR embl ,
+.BR genbank.
+Alignment formats also work;
+common choices include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+For more information, and for codes for some less common formats,
+see main documentation.
+The string
+.I <s>
+is case-insensitive (\fBfasta\fR or \fBFASTA\fR both work).
+
+
+.TP
+.BI \-\-seed " <n>"
+Specify the seed for the random number generator, where the seed
+.I <n>
+is an integer greater than zero. This can be used to make the results of 
+.B esl\-shuffle 
+reproducible.
+If 
+.I <n>
+is 0, the random number generator is seeded arbitrarily and
+stochastic simulations will vary from run to run.
+Arbitrary seeding (0) is the default.
+
+
+
+
+.SH SEE ALSO
+
+.nf
+http://bioeasel.org/
+.fi
+
+.SH COPYRIGHT
+
+.nf 
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi 
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi
+

flax_model/alphafold3/_tools/hmmer/share/man/man1/esl-ssdraw.1

@@ -0,0 +1,882 @@
+.TH "esl\-ssdraw" 1  "Aug 2023" "Easel 0.49" "Easel Manual"
+
+.SH NAME
+esl\-ssdraw \- create postscript secondary structure diagrams
+
+.SH SYNOPSIS
+.B esl\-ssdraw
+[\fIoptions\fR]
+.I msafile
+.I postscript_template
+.I postscript_output_file
+
+.SH DESCRIPTION
+
+.PP
+.B esl\-ssdraw
+reads an existing template consensus secondary structure diagram from
+.I postscript_template
+and creates new postscript diagrams including the template structure
+but with positions colored differently based on alignment statistics
+such as frequency of gaps per position, average posterior probability
+per position or information content per position. Additionally, all
+or some of the aligned sequences can be drawn separately, with
+nucleotides or posterior probabilities mapped onto the corresponding
+positions of the consensus structure.
+
+.PP
+The alignment must be in Stockholm format with per-column reference
+annotation (#=GC RF). The sequences in the alignment must be RNA or
+DNA sequences. The
+.I postscript_template
+file must contain one page that includes <rflen> consensus nucleotides
+(positions), where <rflen> is the number of nongap characters in the
+reference (RF) annotation of the first alignment in
+.IR msafile .
+The specific format required in the 
+.I postscript_template
+is described below in the INPUT section.
+Postscript diagrams will only be created for the first alignment in
+.IR msafile . 
+
+
+.SH OUTPUT
+
+.PP
+By default (if run with zero command line options),
+.B esl\-ssdraw
+will create a six or seven page 
+.IR postscript_output_file , 
+with each page displaying a different alignment statistic. These pages display
+the alignment consensus sequence, information content per position,
+mutual information per position, frequency of inserts per position,
+average length of inserts per position, frequency of
+deletions (gaps) per position, and average posterior probability per
+position (if posterior probabilites exist in the alignment)
+If 
+.B \-d 
+is enabled, all of these pages plus additional ones, such as
+individual sequences (see discussion of 
+.B \-\-indi 
+below) will be drawn.
+These pages can be selected to be drawn individually by using 
+the command line options
+.BR \-\-cons ,
+.BR \-\-info ,
+.BR \-\-mutinfo ,
+.BR \-\-ifreq ,
+.BR \-\-iavglen ,
+.BR \-\-dall ,
+and
+.BR \-\-prob .
+The calculation of the statistics for each of these options is
+discussed below in the description for each option.
+Importantly, only so-called 'consensus' positions of the alignment
+will be drawn. A consensus position is one that is a nongap nucleotide in
+the 'reference' annotation of the Stockholm alignment (#=GC RF) read
+from
+.IR msafile .
+
+.PP
+By default, a consensus sequence for the input alignment will be
+calculated and displayed on the alignment statistic diagrams. The
+consensus sequence is defined as the most common nucleotide at each 
+consensus position of the alignment. The consensus sequence will not
+be displayed if the 
+.B \-\-no\-cnt
+option is used. The 
+.BR \-\-cthresh ,
+.BR \-\-cambig , 
+and 
+.B \-\-athresh
+options affect the definition of the consensus sequence as
+explained below in the descriptions for those options.
+
+.PP
+If the 
+.BI \-\-tabfile " <f>"
+option is used, a tab-delimited text file 
+.I <f>
+will be created that includes per-position lists of the numerical
+values for each of the calculated statistics that were drawn to 
+.IR postscript_output_file .
+Comment lines in
+.I <f>
+are prefixed with a '#' character and explain the meaning of
+each of the tab-delimited columns and how each of the statistics was
+calculated.
+
+If 
+.B \-\-indi
+is used,
+.B esl\-ssdraw
+will create diagrams showing each sequence in the alignment on a
+separate page, with aligned nucleotides in their corresponding
+position in the structure diagram.  By default, basepaired nucleotides
+will be colored based on their basepair type: either Watson-Crick
+(A:U, U:A, C:G, or G:C), G:U or U:G, or non-canonical (the other ten
+possible basepairs). This coloring can be turned off with the
+.B \-\-no\-bp
+option.
+Also by default, nucleotides that differ from the most common
+nucleotide at each aligned consensus position will be outlined. If
+the most common nucleotide occurs in more than 75% of sequences that
+do not have a gap at that position, the outline will be
+bold. Outlining can be turned off with the 
+.B \-\-no\-ol 
+option.
+
+.PP
+With 
+.BR \-\-indi ,
+if the alignment
+contains posterior probability annotation (#=GR PP), the 
+.I postscript_output_file
+will contain an additional page for each sequence drawn with positions
+colored by the posterior probability of each aligned nucleotide.
+No posterior probability pages will be drawn if the 
+.B \-\-no\-pp
+option is used. 
+
+.PP
+.B esl\-ssdraw
+can also be used to draw 'mask' diagrams which color positions of the
+structure one of two colors depending on if they are included or
+excluded by a mask. This is enabled with the 
+.BI \-\-mask\-col " <f>"
+option. 
+.I <f> 
+must contain a single line of <rflen> characters, where <rflen> is the
+the number of nongap RF characters in the alignment. The line must
+contain only '0' and '1' characters. A '0' at position <x> of the
+string indicates position <x> is excluded from the mask, and a '1'
+indicates position <x> is included by the mask.
+A page comparing the overlap of the 
+.I <f> 
+mask from 
+.B \-\-mask\-col
+and another mask in 
+.I <f2> 
+will be created if the 
+.BI \-\-mask\-diff " <f2>" 
+option is used.
+
+.PP
+If the 
+.BI \-\-mask " <f>"
+option is used, positions excluded by the mask in 
+.I <f>
+will be drawn differently (as open circles by default) than positions
+included by the mask. The style of the masked positions can be
+modified with the 
+.BR \-\-mask\-u ,
+.BR \-\-mask\-x ,
+and 
+.B \-\-mask\-a
+options. 
+
+.PP
+Finally, two different types of input files can be used to customize
+output diagrams using the
+.B \-\-dfile
+and
+.B \-\-efile
+options, as described below.
+
+
+
+.SH INPUT
+
+.PP
+The 
+.I postscript_template_file
+is a postscript file that must be in a very specific format in order for
+.B esl\-ssdraw 
+to work. The specifics of the format, described below, are likely to change in future
+versions of 
+.BR esl\-ssdraw .
+The 
+.I postscript_output_file
+files generated by 
+.B esl\-ssdraw
+will not be valid 
+.I postscript_template_file
+format (i.e. an output file from 
+.B esl\-ssdraw
+cannot be used as an 
+.I postscript_template_file
+in a subsequent run of the program).
+
+.PP
+An example 
+.I postscript_template_file
+('trna\-ssdraw.ps') is included with the Easel distribution in
+the 'testsuite/' subdirectory of the top-level 'easel' directory.
+
+.PP
+The
+.I postscript_template_file
+is a valid postscript file. It includes postscript commands for
+drawing a secondary structure. The commands specify x and y
+coordinates for placing each nucleotide on the page. The 
+.I postscript_template_file
+might also contain commands for drawing lines connecting basepaired
+positions and tick marks indicating every tenth position, though these
+are not required, as explained below. 
+
+.PP
+If you are unfamiliar with the postscript language, it may be useful
+for you to know that a postscript page is, by default, 612 points wide
+and 792 points tall.
+The (0,0) coordinate of a postscript
+file is at the bottom left corner of the page, (0,792) is the top left,
+(612,0) is the bottom right, and (612,792) is the top right. 
+.B esl\-ssdraw
+uses 8 point by 8 point cells for drawing positions of the consensus
+secondary structure. The 'scale' section of the
+.I postscript_template_file
+allows for different 'zoom levels', as described below.
+Also, it is important to know that postscript lines beginning with '%'
+are considered comments and do not include postscript commands.
+
+.PP
+An 
+.B esl\-ssdraw
+.I postscript_template_file
+contains n >= 1 pages, each specifying a consensus secondary structure
+diagram. Each page is delimited by a 'showpage' line in an 'ignore'
+section (as described below).
+.B esl\-ssdraw
+will read all pages of the 
+.I postscript_template_file
+and then choose the appropriate one that corresponds with the
+alignment in 
+.I msafile 
+based on the consensus (nongap RF) length of the alignment. 
+For an alignment of consensus length <rflen>, the first
+page of
+.I postscript_template_file
+that has a structure diagram with consensus length <rflen> will be used
+as the template structure for the alignment.
+
+.PP
+Each page of 
+.I postscript_template_file
+contains blocks of text organized into seven different possible
+sections. Each section must begin with a single line '% begin
+<sectionname>' and end with a single line '% end <sectionname>' and
+have n >= 1 lines in between. On the begin and end lines, there must
+be at least one space between the '%' and the 'begin'
+or 'end'. <sectionname> must be one of the
+following: 'modelname', 'legend', 'scale', 'regurgitate', 'ignore', 'text
+positiontext', 'text nucleotides', 'lines positionticks', or 'lines
+bpconnects'. The n >=1 lines in between the begin and end lines of
+each section must be in a specific format that differs for each
+section as described below.
+
+.PP
+Importantly, each page must end with an 'ignore' section that includes
+a single line 'showpage' between the begin and end lines. This
+lets 
+.B esl\-ssdraw
+know that a page has ended and another might follow.
+
+.PP
+Each page of a 
+.I postscript_template_file
+must include a single 'modelname' section.
+This section  must include exactly one line in between its
+begin and end lines. This line must begin with a '%' character
+followed by a single space. The remainder of the line will be parsed
+as the model name and will appear on each page of 
+.B postscript_output_file
+in the header section. If the name is more than 16 characters, it will
+be truncated in the output.
+
+.PP
+Each page of a 
+.I postscript_template_file
+must include a single 'legend' section.  This section must include
+exactly one line in between its begin and end lines. This line must be
+formatted as '% <d1> <f1> <f2> <d2> <f3>', where <d1> is an integer
+specifying the consensus position with relation to which the legend
+will be placed; <f1> and <f2> specify the x and y axis offsets for the
+top left corner of the legend relative to the x and y position of
+consensus position <d1>; <d2> specifies the size of a cell in the
+legend and <f3> specifies how many extra points should be between the
+right hand edge of the legend and the end of the page. the offset of
+the right hand end of the legend . For example, the line '% 34
+\-40. \-30. 12 0.' specfies that the legend be placed 40 points to the left
+and 30 points below the 34th consensus position, that cells
+appearing in the legend be squares of size 12 points by 12 points, and
+that the right hand side of the legend flush against the right hand
+edge of the printable page. 
+
+.PP
+Each page of a 
+.I postscript_template_file
+must include a single 'scale' section.  This section must include
+exactly one line in between its begin and end lines. This line must be
+formatted as '<f1> <f2> scale', where <f1> and <f2> are both positive
+real numbers that are identical, for example '1.7 1.7 scale' is valid,
+but '1.7 2.7 scale' is not. This line is a valid postscript command
+which specifies the scale or zoom level on the pages in the output. If
+<f1> and <f2> are '1.0' the default scale is used for which the total
+size of the page is 612 points wide and 792 points tall. A scale of
+2.0 will reduce this to 306 points wide by 396 points tall. A scale of
+0.5 will increase it to 1224 points wide by 1584 points tall. A single
+cell corresponding to one position of the secondary structure is 8
+points by 8 points. For larger RNAs, a scale of less than 1.0 is
+appropriate (for example, SSU rRNA models (about 1500 nt) use a scale
+of about 0.6), and for smaller RNAs, a scale of more than 1.0 might be
+desirable (tRNA (about 70 nt) uses a scale of 1.7). The best way to
+determine the exact scale to use is trial and error.
+
+.PP
+Each page of a 
+.I postscript_template_file
+can include n >= 0 'regurgitate' sections.
+These sections can include any number of lines. 
+The text in this section will not be parsed by
+.B esl\-ssdraw
+but will be included in each page of 
+.I postscript_output_file.
+The format of the lines in this section must therefore be valid
+postscript commands. An example of content that might be in a 
+regurgitate section are commands to draw lines and text annotating the
+anticodon on a tRNA secondary structure diagram.
+
+.PP
+Each page of a 
+.I postscript_template_file
+must include at least 1 'ignore' section.
+One of these sections must include a single line that
+reads 'showpage'. This section should be placed at the end of each
+page of the template file.  
+Other ignore sections can include any number of lines. 
+The text in these section will not be parsed by
+.B esl\-ssdraw
+nor will it be included in each page of 
+.IR postscript_output_file .
+An ignore section can contain comments or postscript commands that
+draw features of the
+.I postscript_template_file
+that are 
+unwanted in the 
+.IR postscript_output_file .
+
+.PP
+Each page of a 
+.I postscript_template_file
+must include a single 'text nucleotides' section. This section must
+include exactly <rflen> lines, indicating that the consensus secondary
+structure has exactly <rflen> nucleotide positions. Each line must be of
+the format '(<c>) <x> <y> moveto show' where <c> is a nucleotide (this
+can be any character actually), and <x> and <y> are the coordinates
+specifying the location of the nucleotide on the page, they should be
+positive real numbers. The best way to determine what these
+coordinates should be is manually by trial and error, by inspecting
+the resulting structure as you add each nucleotide. Note that
+.B esl\-ssdraw
+will color an 8 point by 8 point cell for each position, so nucleotides
+should be placed about 8 points apart from each other.
+
+.PP
+Each page of a 
+.I postscript_template_file
+may or may not include a single 'text positiontext' section. This section
+can include n >= 1 lines, each specifying text to be placed next to
+specific positions of the structure, for example, to number them.
+Each line must be of
+the format '(<s>) <x> <y> moveto show' where <s> is a string of text
+to place at coordinates (<x>,<y>) of the postscript page. 
+Currently, the best way to determine what these coordinates is
+manually by trial
+and error, by inspecting the resulting diagram as you add
+each line.
+
+.PP
+Each page of a 
+.I postscript_template_file
+may or may not include a single 'lines positionticks' section. This section
+can include n >= 1 lines, each specifying the location of a tick mark
+on the diagram. Each line must be of
+the format '<x1> <y1> <x2> <y2> moveto show'. A tick mark (line of
+width 2.0) will be drawn from point (<x1>,<y1>) to point (<x2>,<y2>)
+on each page of
+.I postscript_output_file.
+Currently, the best way to determine what these coordinates should be
+is manually by trial and error, by inspecting the resulting diagram as
+you add each line.
+
+.PP
+Each page of a 
+.I postscript_template_file
+may or may not include a single 'lines bpconnects' section. This section
+must include <nbp> lines, where <nbp> is the number of basepairs in
+the consensus structure of the input
+.I msafile
+annotated as #=GC SS_cons. Each line should connect two basepaired
+positions in the consensus structure diagram.
+Each line must be of
+the format '<x1> <y1> <x2> <y2> moveto show'. A line
+will be drawn from point (<x1>,<y1>) to point (<x2>,<y2>)
+on each page of
+.I postscript_output_file.
+Currently, the best way to determine what these coordinates should be
+is manually by trial and error, by inspecting the resulting diagram as
+you add each line. 
+
+
+
+.SH REQUIRED MEMORY 
+
+.PP
+The memory required by 
+.B esl\-ssdraw
+will be equal to roughly the larger of 2 Mb and 
+the size of the first alignment in
+.IR msafile .
+If the 
+.B \-\-small 
+option is used, the memory required will be independent of the
+alignment size. To use 
+.B \-\-small
+the alignment must be in Pfam format, a non-interleaved (1 line/seq)
+version of Stockholm format. 
+
+If the 
+.B \-\-indi
+option is used,
+the required memory may exceed the
+size of the alignment by up to ten-fold, and the output
+.B postscript_output_file 
+may be up to 50 times larger than the
+.B msafile.
+
+.SH OPTIONS
+
+.TP
+.B \-h 
+Print brief help;  includes version number and summary of
+all options, including expert options.
+
+.TP
+.B \-d 
+Draw the default set of alignment summary diagrams: consensus
+sequence, information content, mutual information, insert frequency,
+average insert length, deletion frequency, and average posterior
+probability (if posterior probability annotation exists in the
+alignment). These diagrams are also drawn by default (if zero command line
+options are used), but using the
+.B \-d 
+option allows the user to add additional pages, such as individual
+aligned sequences with
+.BR \-\-indi .
+
+.TP 
+.BI \-\-mask " <f>"
+Read the mask from file
+.IR <f> ,
+and draw positions differently in 
+.I postscript_output_file
+depending on whether they are included or excluded by the mask.
+.I <f>
+must contain a single line of length <rflen> with only '0' and '1'
+characters. <rflen> is the number of nongap characters in the
+reference (#=GC RF) annotation of the first alignment in 
+.I msafile
+A '0' at position <x> of the mask indicates position <x> is excluded
+by the mask, and a '1' indicates that position <x> is included by the mask.
+
+.TP 
+.B \-\-small
+Operate in memory saving mode. Without
+.BR \-\-indi ,
+required RAM will be independent of the
+size of the alignment in 
+.IR msafile .
+With
+.BR \-\-indi , 
+the required RAM will be roughly ten times the size of the alignment
+in 
+.IR msafile .
+For 
+.B \-\-small
+to work, the alignment must be in
+Pfam Stockholm (non-interleaved 1 line/seq) format.
+
+.TP 
+.B \-\-rf
+Add a page to 
+.I postscript_output_file 
+showing the reference sequence from the #=GC RF annotation in 
+.I msafile. 
+By default, basepaired nucleotides will be colored based on what type of
+basepair they are. To turn this off, use
+.B \-\-no\-bp.
+This page is drawn by default (if zero command-line options are used).
+
+.TP 
+.B \-\-info
+Add a page to
+.I postscript_output_file
+with consensus (nongap RF) positions colored based on their
+information content from the alignment. 
+Information content is calculated as 2.0 \- H, where H = sum_x p_x
+log_2 p_x for x in {A,C,G,U}. 
+This page is drawn by default (if zero command-line options are used).
+
+.TP 
+.B \-\-mutinfo
+Add a page to
+.I postscript_output_file
+with basepaired consensus (nongap RF) positions colored based on the
+amount of mutual information they have in the alignment. Mutual
+information is sum_{x,y} p_{x,y} log_2 ((p_x * p_y) / p_{x,y}, where x
+and y are the four possible bases A,C,G,U. p_x is the fractions of
+aligned sequences that have nucleotide x of in the left half (5' half) of
+the basepair. p_y is the fraction of aligned sequences that have
+nucleotide y in the position corresponding to the right half (3' half) of
+the basepair. And p_{x,y} is the fraction of aligned sequences that 
+have basepair x:y. For all p_x, p_y and p{x,y} only sequences that 
+that have a nongap nucleotide at both the left and right half of the
+basepair are counted. 
+This page is drawn by default (if zero command-line options are used).
+
+.TP 
+.B \-\-ifreq
+Add a page to
+.I postscript_output_file
+with each consensus (nongap RF) position colored based on the fraction of
+sequences that span each position that have at least 1 inserted
+nucleotide after the position. 
+A sequence s spans consensus position x that is actual alignment
+position a if s has at least one nongap nucleotide aligned to a position
+b <= a and at least one nongap nucleotide aligned to a consensus position
+c >= a. This page is drawn by default (if zero command-line options
+are used).
+
+.TP 
+.B \-\-iavglen
+Add a page to
+.I postscript_output_file
+with each consensus (nongap RF) position colored based on average
+length of insertions that occur after it. The average is calculated as
+the total number of inserted nucleotides after position x, divided by the
+number of sequences that have at least 1 inserted nucleotide after
+position x (so the minimum possible average insert length is 1.0).
+
+.TP 
+.B \-\-dall
+Add a page to
+.I postscript_output_file
+with each consensus (nongap RF) position colored based on the fraction of
+sequences that have a gap (delete) at the position.
+This page is drawn by default (if zero command-line options are used).
+
+.TP 
+.B \-\-dint
+Add a page to
+.I postscript_output_file
+with each consensus (nongap RF) position colored based on the fraction of
+sequences that have an internal gap (delete) at the position. An
+internal gap in a sequence is one that occurs after (5' of) the
+sequence's first aligned nucleotide and after
+(3' of) the sequence's final aligned nucleotide.
+This page is drawn by default (if zero command-line options are used).
+
+.TP 
+.B \-\-prob
+Add a page to
+.I postscript_output_file
+with positions colored based on average posterior probability (PP). The alignment
+must contain #=GR PP annotation for all sequences. PP annotation is
+converted to numerical PP values as follows: '*' = 0.975, '9' =
+0.90, '8' = 0.80, '7' = 0.70, '6' = 0.60, '5' = 0.50, '4' = 0.40, '3'
+= 0.30, '2' = 0.20, '1' = 0.10, '0' = 0.025.
+This page is drawn by default (if zero command-line options are used).
+
+.TP 
+.B \-\-span
+Add a page to
+.I postscript_output_file
+with consensus (nongap RF) positions colored based on the
+fraction of sequences that 'span' the position. 
+A sequence s spans consensus position x that is actual alignment
+position a if s has at least one nongap nucleotide aligned to a position
+b <= a and at least one nongap nucleotide aligned to a consensus position
+c >= a. This page is drawn by default (if zero command-line options
+are used).
+
+
+.SH OPTIONS FOR DRAWING INDIVIDUAL ALIGNED SEQUENCES
+
+.TP 
+.B \-\-indi
+Add a page displaying the aligned nucleotides in their corresponding
+consensus positions of the structure diagram for each aligned
+sequence in the alignment. 
+By default, basepaired nucleotides will be colored based on what type of
+basepair they are. To turn this off, use
+.B \-\-no\-bp.
+If posterior probability information (#=GR
+PP) exists in the alignment, one additional page per sequence will be
+drawn displaying the posterior probabilities.
+
+.TP
+.B \-f
+With 
+.BR \-\-indi ,
+force 
+.B esl\-ssdraw
+to create a diagram, even if it is predicted to be large (> 100 Mb).
+By default, if the predicted size exceeds 100 Mb, 
+.B esl\-ssdraw
+will fail with a warning. 
+
+
+.SH OPTIONS FOR OMITTING PARTS OF THE DIAGRAMS
+
+.TP
+.B \-\-no\-leg
+Omit the legend on all pages of 
+.IR postscript_output_file .
+
+.TP
+.B \-\-no\-head
+Omit the header on all pages of 
+.IR postscript_output_file .
+
+.TP
+.B \-\-no\-foot
+Omit the footer on all pages of 
+.IR postscript_output_file .
+
+
+
+.SH OPTIONS FOR SIMPLE TWO-COLOR MASK DIAGRAMS
+
+.TP
+.B \-\-mask\-col 
+With
+.BR \-\-mask ,
+.I postscript_output_file
+will contain exactly 1 page showing positions included by the mask as 
+black squares, and positions excluded as pink squares.
+
+.TP
+.BI \-\-mask\-diff " <f>"
+With
+.BI \-\-mask  " <f2>"
+and
+.BR mask\-col ,
+.I postscript_output_file
+will contain one additional page comparing the mask from 
+.I <f>
+and the mask from
+.IR <f2> .
+Positions will be colored based on whether they are included by one
+mask and not the other, excluded by both masks, and included by both
+masks.
+
+
+.SH EXPERT OPTIONS FOR CONTROLLING INDIVIDUAL SEQUENCE DIAGRAMS
+
+.TP
+.B \-\-no\-pp
+When used in combination with 
+.BR \-\-indi ,
+do not draw posterior probability structure diagrams for each
+sequence, even if the alignment has PP annotation.
+
+.TP
+.B \-\-no\-bp
+Do not color basepaired nucleotides based on their basepair type.
+
+.TP
+.B \-\-no\-ol
+When used in combination with 
+.BR \-\-indi ,
+do not outline nucleotides that differ from the majority rule
+consensus nucleotide given the alignment.
+
+.TP
+.B \-\-no\-ntpp
+When used in combination with 
+.BR \-\-indi ,
+do not draw nucleotides on the individual sequence posterior
+probability diagrams.
+
+
+.SH EXPERT OPTIONS RELATED TO CONSENSUS SEQUENCE DEFINITION
+
+.TP
+.B \-\-no\-cnt
+Do not draw consensus nucleotides on alignment statistic diagrams (such as
+information content diagrams). By default, the consensus nucleotide is
+defined as the most frequent nucleotide in the alignment at the
+corresponding position. Consensus nucleotides that occur in at least
+.I <x>
+fraction of the aligned sequences (that do not contain a gap at the
+position) are capitalized. By default 
+.I <x>
+is 0.75, but can be changed with the 
+.BI \-\-cthresh " <x>"
+option. 
+
+.TP
+.BI \-\-cthresh " <x>"
+Specify the threshold for capitalizing consensus nucleotides defined by
+the majority rule (i.e. when 
+.B \-\-cambig
+is not enabled) as 
+.IR <x> .
+
+.TP
+.B \-\-cambig
+Change how consensus nucleotides are calculated from majority rule to
+the least ambiguous IUPAC nucleotide that represents at least
+.I <x>
+fraction of the nongap nucleotides at each consensus position. 
+By default 
+.I <x>
+is 0.9, but can be changed with the 
+.BI \-\-athresh " <x>"
+option. 
+
+.TP
+.BI \-\-athresh " <x>"
+With
+.BR \-\-cambig ,
+specify the threshold for defining consensus nucleotides
+is the least ambiguous IUPAC nucleotide that represents at least
+.I <x>
+fraction of the nongap nucleotides at each position.
+
+
+.SH EXPERT OPTIONS CONTROLLING STYLE OF MASKING POSITIONS
+
+.TP
+.B \-\-mask\-u
+With 
+.BR \-\-mask ,
+change the style of masked columns to squares.
+
+.TP
+.B \-\-mask\-x
+With 
+.BR \-\-mask , 
+change the style of masked columns to x's.
+
+.TP
+.B \-\-mask\-a
+With 
+.B \-\-mask
+and
+.B \-\-mask\-u
+or
+.B \-\-mask\-x
+draw the alternative style of square or 'x' masks.
+
+
+.SH EXPERT OPTIONS RELATED TO INPUT FILES
+
+.TP
+.BI \-\-dfile " <f>"
+Read the 'draw file'
+.I <f>
+which specifies numerical values for each consensus position in one or
+more postscript pages.  For each page, the draw file must include
+<rflen>+3 lines (<rflen> is defined in the DESCRIPTION section). The first three lines are special. The following
+<rflen> 'value lines' each must contain a single number, the numerical
+value for the corresponding position.  The first of the three special
+lines defines the 'description' for the page. This should be text that
+describes what the numerical values refer to for the page. The maximum
+allowable length is roughly 50 characters (the exact maximum
+length depends on the template file and the program will report an
+informative error message upon execution if it is exceeded). The
+second special line defines the 'legend header' line that which will
+appear immediately above the legend. It has a maximum allowable length
+of about 30 characters.  The third special line per page must contain
+exactly 7 numbers, which must be in increasing order, each separated
+by a space. 
+These numbers
+define the numerical ranges for the six different colors used to draw
+the consensus positions on the page. 
+The first number defines the minimum value for the first color
+(blue) and must be less than or equal to the minimum
+value from the value lines. The second number defines the minimum
+value for the second color (turquoise). The third, fourth, fifth and
+sixth numbers define the minimum values for the third, fourth, fifth
+and sixth colors (light green, yellow, orange, red), and the seventh
+final number defines the maximum value for red and must be equal to
+or greater than the maximum value from the value lines. 
+After the <rflen> value lines, there must exist a special line with
+only '//', signifying the end of a page.
+The draw file 
+.I <f>
+must end with this special '//' line, even if it only includes a
+single page. A draw file specifying <n> pages should include exactly
+<n> * (<rflen> + 4) lines.
+
+.TP
+.BI \-\-efile " <f>"
+Read the 'expert draw file'
+.I <f>
+which specifies the colors and nucleotides to draw on each consensus
+position in one or more postscript pages. Unlike with the 
+.B \-\-dfile
+option, no legend will be drawn when
+.B \-\-efile 
+is used.
+For each page, the draw file must include <rflen> lines, each with four
+or five tab-delimited tokens. The first four tokens on line <x>
+specify the color to paint position <x> and must be real numbers
+between 0 and 1. The four numbers specify the cyan, magenta, yellow
+and black values, respectively, in the CMYK color scheme for the
+postscript file. The fifth token on line <x> specifies which nucleotide
+to write on position <x> (on top of the colored background). If the
+fifth token does not exist, no nucleotide will be written. 
+After the <rflen> lines, there must exist a special line with
+only '//', signifying the end of a page.
+The expert draw file 
+.I <f>
+must end with this special '//' line, even if it only includes a
+single page. A expert draw file specifying <n> pages should include exactly
+<n> * (<rflen> + 1) lines.
+
+.TP
+.BI \-\-ifile " <f>"
+Read insert information from the file
+.IR <f> ,
+which may have been created with INFERNAL's
+.BR cmalign (1)
+program. The insert information in 
+.I msafile
+will be ignored and the information from
+.I <f>
+will supersede it. Inserts are columns that are gaps in the reference
+(#=GC RF) annotation. 
+
+
+
+
+.SH SEE ALSO
+
+.nf
+http://bioeasel.org/
+.fi
+
+.SH COPYRIGHT
+
+.nf 
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi 
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi

flax_model/alphafold3/_tools/hmmer/share/man/man1/esl-translate.1

@@ -0,0 +1,314 @@
+.TH "esl\-translate" 1  "Aug 2023" "Easel 0.49" "Easel Manual"
+
+.SH NAME
+esl\-translate \- translate DNA sequence in six frames into individual ORFs
+
+.SH SYNOPSIS
+.B esl\-translate
+[\fIoptions\fR]
+.I seqfile
+
+
+.SH DESCRIPTION
+
+.PP
+Given a 
+.I seqfile 
+containing DNA or RNA sequences,
+.B esl\-translate
+outputs a six-frame translation of them as individual open reading
+frames in FASTA format.
+
+.PP
+By default, only open reading frames greater than 20aa are reported. 
+This minimum ORF length can be changed with the
+.B \-l 
+option.
+
+.PP
+By default, no specific initiation codon is required, and any amino acid can start an open reading frame.
+This is so
+.B esl\-translate
+may be used on sequence fragments, eukaryotic genes with introns, or other
+cases where we
+do not want to assume that ORFs are complete coding regions.
+This behavior can be changed. With the 
+.B \-m 
+option, ORFs start with an initiator AUG Met. With the
+.B \-M
+option, ORFs start with any of the initiation codons allowed by the
+genetic code. For example, the "standard" code (NCBI transl_table 1) 
+allows AUG, CUG, and UUG as initiators. When
+.B \-m
+or
+.B \-M
+are used, an initiator is always translated to Met (even if the initiator
+is something like UUG or CUG that doesn't encode Met as an elongator).
+
+.PP
+If
+.I seqfile
+is \- (a single dash), input is read from the stdin pipe. This
+(combined
+with the output being a standard FASTA file) allows
+.B esl\-translate 
+to be used in command line incantations.
+If
+.I seqfile
+ends in .gz, it is assumed to be a gzip-compressed file, and 
+Easel will try to read it as a stream from
+\fBgunzip \-c\fR.
+
+
+
+.SH OUTPUT FORMAT
+
+.PP
+The output FASTA name/description line contains information about the
+source and coordinates of each ORF. Each ORF is named 
+.B orf1,
+etc., with numbering starting from 1, in order of their start position
+on the top strand followed by the bottom strand.  The rest of the
+FASTA name/desc line contains 4 additional fields, followed by the
+description of the source sequence:
+
+.TP
+\fBsource\fR=\fI<s>\fR
+.I <s>
+is the name of the source DNA/RNA sequence.
+
+.TP
+\fBcoords\fR=\fIstart\fR..\fIend\fR
+Coords, 1..L, for the translated ORF in a source DNA sequence of
+length L. If start is greater than end, the ORF is on the bottom
+(reverse complement) strand. The start is the first nucleotide of the
+first codon; the end is the last nucleotide of the last codon. The
+stop codon is not included in the coordinates (unlike in CDS
+annotation in GenBank, for example.)
+
+.TP
+\fBlength\fR=\fI<n>\fR
+Length of the ORF in amino acids.
+
+.TP
+\fBframe\fR=\fI<n>\fR
+Which frame the ORF is in. Frames 1..3 are the top strand; 4..6 are the
+bottom strand. Frame 1 starts at nucleotide 1. Frame 4 starts at
+nucleotide L.
+
+
+
+.SH ALTERNATIVE GENETIC CODES
+
+.PP
+By default, the "standard" genetic code is used (NCBI transl_table 1). 
+Any NCBI genetic code transl_table can be selected with the
+.B \-c 
+option, as follows:
+
+.TP 
+.B 1 
+Standard
+.TP
+.B 2 
+Vertebrate mitochondrial
+.TP
+.B 3
+Yeast mitochondrial
+.TP
+.B 4 
+Mold, protozoan, coelenterate mitochondrial; Mycoplasma/Spiroplasma
+.TP
+.B 5 
+Invertebrate mitochondrial
+.TP
+.B 6 
+Ciliate, dasycladacean, Hexamita nuclear
+.TP
+.B  9 
+Echinoderm and flatworm mitochondrial
+.TP
+.B 10 
+Euplotid nuclear
+.TP
+.B 11
+Bacterial, archaeal; and plant plastid
+.TP
+.B 12 
+Alternative yeast
+.TP
+.B 13 
+Ascidian mitochondrial
+.TP
+.B 14 
+Alternative flatworm mitochondrial
+.TP
+.B 16 
+Chlorophycean mitochondrial
+.TP
+.B 21 
+Trematode mitochondrial
+.TP
+.B 22 
+Scenedesmus obliquus mitochondrial
+.TP
+.B 23 
+Thraustochytrium mitochondrial
+.TP
+.B 24 
+Pterobranchia mitochondrial
+.TP
+.B 25 
+Candidate Division SR1 and Gracilibacteria
+
+
+.PP
+As of this writing, more information about the genetic codes in the
+NCBI translation tables is at 
+.I http://www.ncbi.nlm.nih.gov/Taxonomy/ 
+at a link titled
+.I Genetic codes.
+
+.SH IUPAC DEGENERACY CODES IN DNA 
+
+.PP
+DNA sequences may contain IUPAC degeneracy codes, such as N, R, Y,
+etc. If all codons consistent with a degenerate codon translate to the
+same amino acid (or to a stop), that translation is done; otherwise,
+the codon is translated as X (even if one or more compatible codons
+are stops). For example, in the standard code, UAR translates to *
+(stop), GGN translates to G (glycine), NNN translates to X, and UGR
+translates to X (it could be either a UGA stop or a UGG Trp).
+
+.PP
+Degenerate initiation codons are handled essentially the same. If all
+codons consistent with the degenerate codon are legal initiators, then
+the codon is allowed to initiate a new ORF. Stop codons are never
+a legal initiator (not only with 
+.B \-m 
+or
+.B \-M
+but also with the default of allowing any amino acid to initiate),
+so degenerate codons consistent with a stop cannot be initiators.
+For example, NNN cannot initiate an ORF, nor can UGR -- even
+though they translate to X. This means that we don't translate
+long stretches of N's as long ORFs of X's, which is probably a
+feature, given the prevalence of artificial runs of N's in genome 
+sequence assemblies.
+
+.PP
+Degenerate DNA codons are not translated to degenerate amino acids
+other than X, even when that is possible. For example, SAR and MUH
+are decoded as X, not Z (Q|E) and J (I|L). The extra complexity
+needed for a degenerate to degenerate translation doesn't seem worthwhile.
+
+
+.SH OPTIONS
+
+.TP
+.B \-h
+Print brief help. Includes version number and summary of all options. 
+Also includes a list of the available
+NCBI transl_tables and their numerical codes, for the
+.B \-c 
+option.
+
+.TP
+.BI \-c " <id>"
+Choose alternative genetic code 
+.I <id>
+where 
+.I <id>
+is the numerical code of one of the NCBI transl_tables.
+
+.TP
+.BI \-l " <n>"
+Set the minimum reported ORF length to 
+.I <n>
+aa.
+
+.TP
+.B \-m
+Require ORFs to start with an initiator codon AUG (Met).
+
+.TP
+.B \-M
+Require ORFs to start with an initiator codon, as specified by the
+allowed initiator codons in the NCBI transl_table. In the default
+Standard code, AUG, CUG, and UUG are allowed as initiators. An 
+initiation codon is always translated as Met, even if it does not
+normally encode Met as an elongator.
+
+.TP
+.B \-W
+Use a memory-efficient windowed sequence reader.
+The default is to read entire DNA sequences into memory, which
+may become memory limited for some very large eukaryotic chromosomes.
+The windowed reader cannot 
+reverse complement a nonrewindable input stream, so 
+either
+.I seqfile
+must be a file,
+or you must use
+.I \-\-watson
+to limit translation to the top strand.
+
+
+.TP
+.BI \-\-informat " <s>"
+Assert that input
+.I seqfile
+is in format
+.IR <s> ,
+bypassing format autodetection.
+Common choices for 
+.I <s> 
+include:
+.BR fasta ,
+.BR embl ,
+.BR genbank.
+Alignment formats also work;
+common choices include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+For more information, and for codes for some less common formats,
+see main documentation.
+The string
+.I <s>
+is case-insensitive (\fBfasta\fR or \fBFASTA\fR both work).
+
+
+.TP
+.B \-\-watson
+Only translate the top strand.
+
+.TP
+.B \-\-crick
+Only translate the bottom strand.
+
+
+
+
+
+.SH SEE ALSO
+
+.nf
+http://bioeasel.org/
+.fi
+
+.SH COPYRIGHT
+
+.nf 
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi 
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi

flax_model/alphafold3/_tools/hmmer/share/man/man1/esl-weight.1

@@ -0,0 +1,105 @@
+.TH "esl\-weight" 1  "Aug 2023" "Easel 0.49" "Easel Manual"
+
+.SH NAME
+esl\-weight \- calculate sequence weights in MSA(s)
+
+.SH SYNOPSIS
+.B esl\-weight
+[\fIoptions\fR]
+.I msafile
+
+.SH DESCRIPTION
+
+.PP
+.B esl\-weight
+calculates individual sequence weights for each alignment in 
+.I msafile 
+and outputs a new 
+multiple sequence alignment file in Stockholm format with
+the weights annotated in Stockholm-format 
+\fB#=GS \fR\fIseqname\fR\fB WT \fR\fIweight\fR
+lines.
+The default weighting algorithm is the Gerstein/Sonnhammer/Chothia
+algorithm.
+
+.PP
+If 
+.I msafile
+is \- (a single dash),
+MSA input is read from stdin.
+
+
+
+
+.SH OPTIONS
+
+.TP
+.B \-h 
+Print brief help;  includes version number and summary of
+all options, including expert options.
+
+.TP
+.B \-g
+Use the Gerstein/Sonnhammer/Chothia weighting algorithm; this is the
+default.
+
+.TP
+.B \-p
+Use the Henikoff position-based weighting algorithm. This is faster
+and more memory efficient than the default.
+
+.TP
+.B \-b
+"BLOSUM weights": use approximately the same rule used in constructing
+the BLOSUM score matrices. This involves single-linkage clustering at
+some fractional identity threshold (default 0.62; see 
+.B \-\-id 
+option), then for each cluster, splitting a total weight of one
+uniformly amongst all sequences in the cluster.
+
+
+.SH EXPERT OPTIONS
+
+.TP 
+.BI \-\-id " <x>"
+Sets the fractional identity threshold used by the BLOSUM weighting
+rule (option 
+.BR \-b ;
+required), to a number 0<=x<=1. Default is 0.62.
+
+.TP
+.B \-\-amino
+Assert that the 
+.I msafile 
+contains protein sequences. 
+
+.TP 
+.B \-\-dna
+Assert that the 
+.I msafile 
+contains DNA sequences. 
+
+.TP 
+.B \-\-rna
+Assert that the 
+.I msafile 
+contains RNA sequences. 
+
+.SH SEE ALSO
+
+.nf
+http://bioeasel.org/
+.fi
+
+.SH COPYRIGHT
+
+.nf 
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi 
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi

flax_model/alphafold3/_tools/hmmer/share/man/man1/hmmbuild.1

@@ -0,0 +1,556 @@
+.TH "hmmbuild" 1 "Aug 2023" "HMMER 3.4" "HMMER Manual"
+
+.SH NAME
+hmmbuild \- construct profiles from multiple sequence alignments
+
+.SH SYNOPSIS
+.B hmmbuild
+[\fIoptions\fR]
+.I hmmfile
+.I msafile
+
+
+.SH DESCRIPTION
+
+For each multiple sequence alignment in 
+.I msafile
+build a profile HMM 
+and save it to a new file
+.IR hmmfile .
+
+
+.PP
+.I msafile 
+may be '\-' (dash), which means
+reading this input from stdin rather than a file. 
+
+
+.PP
+.I hmmfile
+may not be '\-' (stdout), because sending the HMM file to stdout would
+conflict with the other text output of the program.
+
+
+
+
+.SH OPTIONS
+
+.TP
+.B \-h
+Help; print a brief reminder of command line usage and all available
+options.
+
+.TP
+.BI \-n " <s>"
+Name the new profile 
+.IR <s> .
+The default is to use the name of the alignment (if one is present in 
+the 
+.IR msafile ,
+or, failing that, the name of the
+.IR hmmfile .
+If 
+.I msafile
+contains more than one alignment, 
+.B \-n
+doesn't work, and every alignment must have a name 
+annotated in the 
+.I msafile
+(as in Stockholm #=GF ID annotation).
+
+
+.TP
+.BI \-o " <f>"
+Direct the summary output to file
+.IR <f> ,
+rather than to stdout.
+
+.TP
+.BI \-O " <f>"
+After each model is constructed, resave annotated, possibly modified
+source alignments to a file
+.I <f>
+in Stockholm format.
+The alignments are annotated with a reference annotation line
+indicating which columns were assigned as consensus, and sequences are
+annotated with what relative sequence weights were assigned. Some
+residues of the alignment may have been shifted to accommodate
+restrictions of the Plan7 profile architecture, which disallows
+transitions between insert and delete states.
+
+
+.SH OPTIONS FOR SPECIFYING THE ALPHABET
+
+.TP
+.B \-\-amino
+Assert that sequences in 
+.I msafile
+are protein, bypassing alphabet autodetection.
+
+.TP
+.B \-\-dna
+Assert that sequences in
+.I msafile
+are DNA, bypassing alphabet autodetection.
+
+.TP
+.B \-\-rna
+Assert that sequences in 
+.I msafile
+are RNA, bypassing alphabet autodetection.
+
+.SH OPTIONS CONTROLLING PROFILE CONSTRUCTION 
+
+These options control how consensus columns are defined in an alignment.
+
+.TP
+.B \-\-fast 
+Define consensus columns as those that have a fraction >= 
+.B symfrac
+of residues as opposed to gaps. (See below for the
+.B \-\-symfrac
+option.) This is the default.
+
+.TP
+.B \-\-hand
+Define consensus columns in next profile using reference annotation to
+the multiple alignment. 
+This allows you to define any consensus columns you like.
+
+.TP
+.BI \-\-symfrac " <x>"
+Define the residue fraction threshold necessary to define a
+consensus column when using the 
+.B \-\-fast 
+option. The default is 0.5. The symbol fraction in each column
+is calculated after taking relative sequence weighting into account,
+and ignoring gap characters corresponding to ends of sequence
+fragments
+(as opposed to internal insertions/deletions).
+Setting this to 0.0 means that every alignment column will be assigned
+as consensus, which may be useful in some cases. Setting it to 1.0
+means that only columns that include 0 gaps (internal
+insertions/deletions) will be assigned as consensus.
+
+.TP
+.BI \-\-fragthresh " <x>"
+We only want to count terminal gaps as deletions if the aligned
+sequence is known to be full-length, not if it is a fragment (for
+instance, because only part of it was sequenced). HMMER uses a simple
+rule to infer fragments: if the range of a sequence in the alignment 
+(the number of alignment columns between the first and last positions 
+of the sequence) is less than or equal to a fraction
+.I <x> 
+times the alignment length in columns,
+then the sequence is handled as a fragment. The default is 0.5.
+Setting
+.B \-\-fragthresh 0
+will define no (nonempty) sequence as a fragment; you might want to do
+this if you know you've got a carefully curated alignment of full\-length
+sequences.
+Setting
+.B \-\-fragthresh 1
+will define all sequences as fragments; you might want to do this if
+you know your alignment is entirely composed of fragments, such as
+translated short reads in metagenomic shotgun data.
+
+
+.SH OPTIONS CONTROLLING RELATIVE WEIGHTS
+
+HMMER uses an ad hoc sequence weighting algorithm to downweight
+closely related sequences and upweight distantly related ones. This
+has the effect of making models less biased by uneven phylogenetic
+representation. For example, two identical sequences would typically
+each receive half the weight that one sequence would.  These options
+control which algorithm gets used.
+
+.TP
+.B \-\-wpb
+Use the Henikoff position-based sequence weighting scheme [Henikoff
+and Henikoff, J. Mol. Biol. 243:574, 1994].  This is the default.
+
+.TP 
+.B \-\-wgsc 
+Use the Gerstein/Sonnhammer/Chothia weighting algorithm [Gerstein et
+al, J. Mol. Biol. 235:1067, 1994].
+
+.TP 
+.B \-\-wblosum
+Use the same clustering scheme that was used to weight data in
+calculating BLOSUM substitution matrices [Henikoff and Henikoff,
+Proc. Natl. Acad. Sci 89:10915, 1992]. Sequences are single-linkage
+clustered at an identity threshold (default 0.62; see
+.BR \-\-wid )
+and within each cluster of c sequences, each sequence gets relative
+weight 1/c.
+
+.TP
+.B \-\-wnone
+No relative weights. All sequences are assigned uniform weight. 
+
+.TP 
+.BI \-\-wid " <x>"
+Sets the identity threshold used by single-linkage clustering when 
+using 
+.BR \-\-wblosum . 
+Invalid with any other weighting scheme. Default is 0.62.
+
+
+
+
+.SH OPTIONS CONTROLLING EFFECTIVE SEQUENCE NUMBER
+
+After relative weights are determined, they are normalized to sum to a
+total effective sequence number, 
+.IR eff_nseq . 
+This number may be the actual number of sequences in the alignment,
+but it is almost always smaller than that.
+The default entropy weighting method 
+(\fB\-\-eent\fR)
+reduces the effective sequence
+number to reduce the information content (relative entropy, or average
+expected score on true homologs) per consensus position. The target
+relative entropy is controlled by a two-parameter function, where the
+two parameters are settable with
+.B \-\-ere
+and 
+.BR \-\-esigma .
+
+.TP
+.B \-\-eent
+Adjust effective sequence number to achieve a specific relative entropy
+per position (see
+.BR \-\-ere ).
+This is the default.
+
+.TP
+.B \-\-eclust
+Set effective sequence number to the number of single-linkage clusters
+at a specific identity threshold (see 
+.BR \-\-eid ).
+This option is not recommended; it's for experiments evaluating
+how much better
+.B \-\-eent
+is.
+
+.TP
+.B \-\-enone
+Turn off effective sequence number determination and just use the
+actual number of sequences. One reason you might want to do this is
+to try to maximize the relative entropy/position of your model, which
+may be useful for short models.
+
+.TP
+.BI \-\-eset " <x>"
+Explicitly set the effective sequence number for all models to 
+.IR <x> .
+
+.TP
+.BI \-\-ere " <x>"
+Set the minimum relative entropy/position target to 
+.IR <x> .
+Requires
+.BR \-\-eent . 
+Default depends on the sequence alphabet. For protein
+sequences, it is 0.59 bits/position; for nucleotide 
+sequences, it is 0.45 bits/position.
+
+.TP
+.BI \-\-esigma " <x>"
+Sets the minimum relative entropy contributed by an entire
+model alignment, over its whole length. This has the effect
+of making short models have 
+higher relative entropy per position than 
+.B \-\-ere 
+alone would give. The default is 45.0 bits.
+
+.TP
+.BI \-\-eid " <x>"
+Sets the fractional pairwise identity cutoff used by 
+single linkage clustering with the
+.B \-\-eclust 
+option. The default is 0.62.
+
+
+.SH OPTIONS CONTROLLING PRIORS
+
+By default, weighted counts are converted to mean posterior
+probability parameter estimates using mixture Dirichlet priors.
+Default mixture Dirichlet prior parameters for protein models and for
+nucleic acid (RNA and DNA) models are built in. The following options
+allow you to override the default priors.
+
+.TP
+.B \-\-pnone
+Don't use any priors. Probability parameters will simply be the
+observed frequencies, after relative sequence weighting. 
+
+.TP
+.B \-\-plaplace
+Use a Laplace +1 prior in place of the default mixture Dirichlet
+prior.
+
+
+
+
+.SH OPTIONS CONTROLLING SINGLE SEQUENCE SCORING
+
+By default, if a query is a single sequence from a file in 
+.IR fasta
+format,
+.B hmmbuild 
+constructs a search model from that sequence and a standard
+20x20 substitution matrix for residue probabilities, along with two
+additional parameters for position-independent gap open and gap extend
+probabilities. These options allow the default single-sequence scoring
+parameters to be changed, and for single-sequence scoring options to
+be applied to a single sequence coming from an aligned format.
+
+.TP
+.BI \-\-singlemx 
+If a single sequence query comes from a multiple sequence alignment file, 
+such as in 
+.IR stockholm
+format, the search model is by default constructed as is typically done 
+for multiple sequence alignments. This option forces 
+.B hmmbuild 
+to use the single-sequence method with substitution score matrix.
+
+.TP
+.BI \-\-mx " <s>"
+Obtain residue alignment probabilities from the built-in
+substitution matrix named
+.IR <s> . 
+Several standard matrices are built-in, and do not need to be
+read from files. 
+The matrix name
+.I <s> 
+can be
+PAM30, PAM70, PAM120, PAM240, BLOSUM45, BLOSUM50, BLOSUM62, BLOSUM80,
+BLOSUM90, or DNA1.
+Only one of the
+.B \-\-mx 
+and
+.B \-\-mxfile
+options may be used.
+
+.TP
+.BI \-\-mxfile " <mxfile>"
+Obtain residue alignment probabilities from the substitution matrix
+in file
+.IR <mxfile> .
+The default score matrix is BLOSUM62 for protein sequences, and 
+DNA1 for nucleotide sequences (these matrices are internal to
+HMMER and do not need to be available as a file). 
+The format of a substitution matrix
+.I <mxfile>
+is the standard format accepted by BLAST, FASTA, and other sequence 
+analysis software. 
+See ftp.ncbi.nlm.nih.gov/blast/matrices/ for example files. (The only
+exception: we require matrices to be square, so for DNA, use files
+like NCBI's NUC.4.4, not NUC.4.2.)
+
+.TP
+.BI \-\-popen " <x>"
+Set the gap open probability for a single sequence query model to 
+.IR <x> .
+The default is 0.02. 
+.I <x> 
+must be >= 0 and < 0.5.
+
+.TP
+.BI \-\-pextend " <x>"
+Set the gap extend probability for a single sequence query model to 
+.IR <x> .
+The default is 0.4. 
+.I <x> 
+must be >= 0 and < 1.0.
+
+
+.SH OPTIONS CONTROLLING E-VALUE CALIBRATION
+
+The location parameters for the expected score distributions for MSV
+filter scores, Viterbi filter scores, and Forward scores require three
+short random sequence simulations.
+
+.TP
+.BI \-\-EmL " <n>"
+Sets the sequence length in simulation that estimates the location
+parameter mu for MSV filter E-values. Default is 200.
+
+.TP
+.BI \-\-EmN " <n>"
+Sets the number of sequences in simulation that estimates the location
+parameter mu for MSV filter E-values. Default is 200.
+
+.TP
+.BI \-\-EvL " <n>"
+Sets the sequence length in simulation that estimates the location
+parameter mu for Viterbi filter E-values. Default is 200.
+
+.TP
+.BI \-\-EvN " <n>"
+Sets the number of sequences in simulation that estimates the location
+parameter mu for Viterbi filter E-values. Default is 200.
+
+.TP
+.BI \-\-EfL " <n>"
+Sets the sequence length in simulation that estimates the location
+parameter tau for Forward E-values. Default is 100.
+
+.TP
+.BI \-\-EfN " <n>"
+Sets the number of sequences in simulation that estimates the location
+parameter tau for Forward E-values. Default is 200.
+
+.TP
+.BI \-\-Eft " <x>"
+Sets the tail mass fraction to fit in the simulation that estimates
+the location parameter tau for Forward evalues. Default is 0.04.
+
+
+.SH OTHER OPTIONS
+
+.TP
+.BI \-\-cpu " <n>"
+Set the number of parallel worker threads to 
+.IR <n> .
+On multicore machines, the default is 2.
+You can also control this number by setting an environment variable, 
+.IR HMMER_NCPU .
+There is also a master thread, so the actual number of threads that
+HMMER spawns is
+.IR <n> +1.
+
+This option is not available if HMMER was compiled with POSIX threads
+support turned off.
+
+
+
+.TP 
+.BI \-\-informat " <s>"
+Assert that input
+.I msafile
+is in alignment format
+.IR <s> ,
+bypassing format autodetection.
+Common choices for 
+.I <s> 
+include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+For more information, and for codes for some less common formats,
+see main documentation.
+The string
+.I <s>
+is case-insensitive (\fBa2m\fR or \fBA2M\fR both work).
+
+
+.TP 
+.BI \-\-seed " <n>"
+Seed the random number generator with
+.IR <n> ,
+an integer >= 0. 
+If 
+.I <n> 
+is nonzero, any stochastic simulations will be reproducible; the same
+command will give the same results.
+If 
+.I <n>
+is 0, the random number generator is seeded arbitrarily, and
+stochastic simulations will vary from run to run of the same command.
+The default seed is 42.
+
+
+.TP 
+.BI \-\-w_beta " <x>"
+Window length tail mass.
+The upper bound,
+.IR W ,
+on the length at which nhmmer expects to find an instance of the 
+model is set such that the fraction of all sequences generated
+by the model with length 
+.I ">= W"
+is less than  
+.IR <x> . 
+The default is 1e-7. 
+
+
+
+.TP 
+.BI \-\-w_length " <n>"
+Override the model instance length upper bound,
+.IR W ,
+which is otherwise controlled by
+.BR \-\-w_beta . 
+It should be larger than the model length. The value of 
+.I W
+is used deep in the acceleration pipeline, and modest changes
+are not expected to impact results (though larger values of 
+.I W
+do lead to longer run time). 
+
+
+.TP
+.B \-\-mpi
+Run as a parallel MPI program. Each alignment is assigned to a MPI
+worker node for construction. (Therefore, the maximum parallelization 
+cannot exceed the number of alignments in the input
+.IR msafile .)
+This is useful when building large profile libraries. This option is
+only available if optional MPI capability was enabled at compile-time.
+
+
+.TP 
+.B \-\-stall
+For debugging MPI parallelization: arrest program execution
+immediately after start, and wait for a debugger to attach to the
+running process and release the arrest.
+
+
+.TP 
+.BI \-\-maxinsertlen " <n>"
+Restrict insert length parameterization such that the expected
+insert length at each position of the model is no more than
+.IR <n> . 
+ 
+
+
+
+.SH SEE ALSO 
+
+See 
+.BR hmmer (1)
+for a master man page with a list of all the individual man pages
+for programs in the HMMER package.
+
+.PP
+For complete documentation, see the user guide that came with your
+HMMER distribution (Userguide.pdf); or see the HMMER web page
+(http://hmmer.org/).
+
+
+
+.SH COPYRIGHT
+
+.nf
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi
+
+For additional information on copyright and licensing, see the file
+called COPYRIGHT in your HMMER source distribution, or see the HMMER
+web page 
+(http://hmmer.org/).
+
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi
+

flax_model/alphafold3/_tools/hmmer/share/man/man1/hmmconvert.1

@@ -0,0 +1,106 @@
+.TH "hmmconvert" 1 "Aug 2023" "HMMER 3.4" "HMMER Manual"
+
+.SH NAME
+hmmconvert \- convert profile file to various formats
+
+
+.SH SYNOPSIS
+.B hmmconvert
+[\fIoptions\fR]
+.I hmmfile
+
+
+.SH DESCRIPTION
+
+.PP
+The
+.B hmmconvert 
+utility
+converts an input profile file to different HMMER formats.
+
+.PP
+By default, the input profile can be in any HMMER format, including
+old/obsolete formats from HMMER2, ASCII or binary; the output profile
+is a current HMMER3 ASCII format.
+
+.PP
+.I hmmfile
+may be '\-' (dash), which means reading this input from stdin rather
+than a file.
+
+
+.SH OPTIONS
+
+.TP
+.B \-h
+Help; print a brief reminder of command line usage and all available
+options.
+
+.TP 
+.B \-a
+Output profiles in ASCII text format. This is the default.
+
+.TP 
+.B \-b
+Output profiles in binary format. 
+
+.TP 
+.B \-2
+Output in legacy HMMER2 ASCII text format, in ls (glocal) mode. This
+allows HMMER3 models to be converted back to a close approximation of
+HMMER2, for comparative studies.
+
+.TP 
+.BI \-\-outfmt " <s>"
+Output in a HMMER3 ASCII text format other then the most current one.
+Valid choices for 
+.I <s>
+are
+.B "3/a"
+through
+.BR "3/f" .
+The current format is
+.BR 3/f ,
+and this is the default. The format
+.B 3/b 
+was used in the official HMMER3 release, and the others were used in
+the various testing versions.
+
+
+.SH SEE ALSO 
+
+See 
+.BR hmmer (1)
+for a master man page with a list of all the individual man pages
+for programs in the HMMER package.
+
+.PP
+For complete documentation, see the user guide that came with your
+HMMER distribution (Userguide.pdf); or see the HMMER web page
+(http://hmmer.org/).
+
+
+
+.SH COPYRIGHT
+
+.nf
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi
+
+For additional information on copyright and licensing, see the file
+called COPYRIGHT in your HMMER source distribution, or see the HMMER
+web page 
+(http://hmmer.org/).
+
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi
+
+
+
+"
+

flax_model/alphafold3/_tools/hmmer/share/man/man1/hmmemit.1

@@ -0,0 +1,250 @@
+.TH "hmmemit" 1 "Aug 2023" "HMMER 3.4" "HMMER Manual"
+
+.SH NAME
+hmmemit \- sample sequences from a profile 
+
+.SH SYNOPSIS
+.B hmmemit
+[\fIoptions\fR]
+.I hmmfile
+
+
+.SH DESCRIPTION
+
+.PP
+The 
+.B hmmemit
+program 
+samples (emits) sequences from the profile HMM(s) in
+.IR hmmfile ,
+and writes them to output.
+Sampling sequences may be useful for a variety of purposes, including
+creating synthetic true positives for benchmarks or tests.
+
+.PP
+The default is to sample one unaligned sequence from the core
+probability model, which means that each sequence consists of one
+full-length domain.  Alternatively, with the
+.B \-c
+option, you can emit a simple majority-rule consensus sequence;
+or with the
+.B \-a 
+option, you can emit an alignment (in which case, you probably
+also want to set 
+.B \-N 
+to something other than its default of 1 sequence per model).
+
+.PP
+As another option, with the
+.B \-p
+option you can sample a sequence from a fully configured HMMER search
+profile. This means sampling a `homologous sequence' by HMMER's
+definition, including nonhomologous flanking sequences, local
+alignments, and multiple domains per sequence, depending on the length
+model and alignment mode chosen for the profile.
+
+.PP
+The
+.I hmmfile 
+may contain a library of HMMs, in which case
+each HMM will be used in turn.
+
+.PP
+.I hmmfile 
+may be '\-' (dash), which
+means reading this input from stdin rather than a file.  
+
+
+.SH COMMON OPTIONS
+
+.TP
+.B \-h
+Help; print a brief reminder of command line usage and all available
+options.
+
+
+.TP
+.BI \-o " <f>"
+Direct the output sequences to file
+.IR <f> ,
+rather than to stdout.
+
+.TP
+.BI \-N " <n>"
+Sample
+.I <n>
+sequences per model, rather than just one.
+
+
+
+.SH OPTIONS CONTROLLING WHAT TO EMIT
+
+The default is to sample
+.B N
+sequences from the core model. Alternatively,
+you may choose one (and only one) of the following alternatives.
+
+
+.TP
+.B \-a
+Emit an alignment for each HMM in the 
+.I hmmfile
+rather than sampling unaligned sequences one at a time.
+
+.TP
+.B \-c
+Emit a plurality-rule consensus sequence, instead of sampling a
+sequence from the profile HMM's probability distribution. The
+consensus sequence is formed by selecting the maximum probability
+residue at each match state.
+
+.TP
+.B \-C
+Emit a fancier plurality-rule consensus sequence than the
+.B \-c
+option. If the maximum probability residue has p < 
+.B minl
+show it as a lower case 'any' residue (n or x); if p >= 
+.B minl 
+and < 
+.B minu 
+show it as a lower case residue; and if p >= 
+.B minu
+show it as an upper case residue.
+The default settings of 
+.B minu
+and 
+.B minl 
+are both 0.0, which means
+.B \-C 
+gives the same output as 
+.B \-c 
+unless you also set 
+.B minu
+and
+.B minl 
+to what you want.
+
+.TP
+.B \-p
+Sample unaligned sequences from the implicit search profile, not from
+the core model.  The core model consists only of the homologous states
+(between the begin and end states of a HMMER Plan7 model). The profile
+includes the nonhomologous N, C, and J states, local/glocal and
+uni/multihit algorithm configuration, and the target length model.
+Therefore sequences sampled from a profile may include nonhomologous
+as well as homologous sequences, and may contain more than one
+homologous sequence segment. By default, the profile is in multihit
+local mode, and the target sequence length is configured for L=400. 
+
+
+
+
+.SH OPTIONS CONTROLLING EMISSION FROM PROFILES
+
+These options require that you have set the
+.B \-p
+option.
+
+.TP
+.BI \-L " <n>"
+Configure the profile's target sequence length model to generate a
+mean length of approximately <n> rather than the default of 400.
+
+.TP
+.B \-\-local
+Configure the profile for multihit local alignment.
+
+.TP
+.B \-\-unilocal
+Configure the profile for unihit local alignment (Smith/Waterman).
+
+.TP
+.B \-\-glocal
+Configure the profile for multihit glocal alignment.
+
+.TP
+.B \-\-uniglocal
+Configure the profile for unihit glocal alignment.
+
+
+.SH OPTIONS CONTROLLING FANCY CONSENSUS EMISSION
+
+These options require that you have set the
+.B \-C
+option.
+
+.TP
+.BI \-\-minl " <x>"
+Sets the 
+.B minl
+threshold for showing weakly conserved residues as lower case.
+(0 <= x <= 1)
+
+.TP
+.BI \-\-minu " <x>"
+Sets the 
+.B minu 
+threshold for showing strongly conserved residues as upper case.
+(0 <= x <= 1)
+
+
+
+.SH OTHER OPTIONS
+
+.TP
+.BI \-\-seed " <n>"
+Seed the random number generator with
+.IR <n> ,
+an integer >= 0. 
+If 
+.I <n> 
+is nonzero, any stochastic simulations will be reproducible; the same
+command will give the same results.
+If 
+.I <n>
+is 0, the random number generator is seeded arbitrarily, and
+stochastic simulations will vary from run to run of the same command.
+The default is 0: use an arbitrary seed, so different
+.B hmmemit
+runs will generate different samples.
+
+
+
+
+
+.SH SEE ALSO 
+
+See 
+.BR hmmer (1)
+for a master man page with a list of all the individual man pages
+for programs in the HMMER package.
+
+.PP
+For complete documentation, see the user guide that came with your
+HMMER distribution (Userguide.pdf); or see the HMMER web page
+(http://hmmer.org/).
+
+
+
+.SH COPYRIGHT
+
+.nf
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi
+
+For additional information on copyright and licensing, see the file
+called COPYRIGHT in your HMMER source distribution, or see the HMMER
+web page 
+(http://hmmer.org/).
+
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi
+
+
+

flax_model/alphafold3/_tools/hmmer/share/man/man1/hmmer.1

@@ -0,0 +1,199 @@
+.TH "HMMER" 1 "Aug 2023" "HMMER 3.4" "HMMER Manual"
+
+.SH NAME
+
+HMMER \- profile HMMs for biological sequence analysis
+
+.SH SYNOPSIS
+
+.nf
+.B hmmalign
+  Align sequences to a profile 
+
+.B hmmbuild
+  Construct profiles from multiple sequence alignments
+
+.B hmmconvert
+  Convert profile file to various formats
+
+.B hmmemit
+  Sample sequences from a profile
+
+.B hmmfetch
+  Retrieve profiles from a file
+
+.B hmmlogo
+  Produce a conservation logo graphic from a profile
+
+.B hmmpgmd
+  Daemon for database search web services
+
+.B hmmpress
+  Prepare a profile database for hmmscan
+
+.B hmmscan
+  Search sequence(s) against a profile database
+
+.B hmmsearch
+  Search profile(s) against a sequence database
+
+.B hmmsim
+  Collect profile score distributions on random sequences
+
+.B hmmstat
+  Summary statistics for a profile file
+
+.B jackhmmer
+  Iteratively search sequence(s) against a sequence database
+
+.B makehmmerdb
+  build nhmmer database from a sequence file
+
+.B nhmmer
+  Search DNA/RNA queries against a DNA/RNA sequence database
+
+.B nhmmscan
+  Search DNA/RNA sequence(s) against a DNA/RNA profile database
+
+.B phmmer
+  Search protein sequence(s) against a protein sequence database
+
+.B alimask
+  Calculate and add column mask to a multiple sequence alignment
+.fi
+
+.SH DESCRIPTION
+
+HMMER is a suite of several programs for biological sequence alignment
+and database homology search. It uses probabilistic models called
+"profile hidden Markov models" (profile HMMs) to represent the likely
+evolutionary homologs of a single sequence or a multiple alignment of
+a sequence family. A main avenue of research is to improve the
+evolutionary predictive models in HMMER to be able to recognize and
+accurately align increasingly remote homologs, distant in time.
+
+HMMER is also used as an organizational tool, to group the
+exponentially growing number of biological sequences into a vastly
+smaller set of well-annotated sequence families. New sequences can be
+annotated by comparison against curated sequence family databases of
+prebuilt HMMER profiles, in addition or instead of comparison to the
+entire sequence database. Databases such as Pfam, SMART, and
+TIGRfams, among others, are based on this principle.
+
+HMMER is used in three main modes: to search a sequence database for
+new homologs of a sequence or a sequence family; to search a profile
+database (like Pfam) to find what known family a query sequence
+belongs to, or what domains it has; and to automatically construct
+large multiple alignments (i.e. with an effectively unlimited number
+of sequences) using a profile representative of a sequence family.
+
+
+Suppose you have a multiple sequence alignment of a sequence family of
+interest, and you want to search a sequence database for additional
+homologs. The
+.B hmmbuild 
+program builds profile(s) from multiple alignment(s). 
+The
+.B hmmsearch
+program searches protein profile(s) against a protein sequence database,
+and 
+.B nhmmer
+searches nucleotide profile(s) against a nucleotide sequence database.
+
+Suppose you have a single sequence of interest, and you want to search
+a sequence database for additional homologs. The
+.B phmmer
+program searches a single protein sequence against a protein sequence
+database. The
+.B jackhmmer 
+program does the same thing but iteratively -- homologs detected in a
+previous round are incorporated into a new profile, and the new
+profile is searched again. 
+.B phmmer 
+is used like BLASTP, and 
+.B jackhmmer
+is used like a protein PSI-BLAST. The
+.B nhmmer
+program searches a single nucleotide sequence against a nucleotide sequence.
+
+Suppose you have sequence(s) that you want to analyze using a
+HMMER-based profile HMM database like Pfam (http://pfam.sanger.ac.uk).
+The
+.B hmmpress
+program formats a profile HMM flatfile (such as the file you
+would download from Pfam) into a HMMER binary database.
+The 
+.B hmmscan
+program searches protein sequence(s) against that database.
+The 
+.B nhmmscan
+program can similarly search nucleotide sequence(s) against
+a pressed database of nucleotide profiles, such as from 
+Dfam (http://dfam.janelia.org).
+
+
+Suppose you want to align lots of sequences. You can construct a
+manageably small alignment of a representative set of sequences,
+build a profile with
+.BR hmmbuild ,
+and use the
+.B hmmalign 
+program to align any number of sequences to that profile.
+
+HMMER also includes some auxiliary tools for working with large
+profile databases.
+.B hmmfetch 
+fetches one or more profiles from a database.
+.B hmmstat 
+prints summary statistics about a profile file.
+
+For compatibility with other profile software and previous versions of
+HMMER, the
+.B hmmconvert
+program converts profiles to a few other formats. We intend to add
+more support for other formats over time.
+
+The
+.B hmmemit 
+program generates (simulates) "homologous" sequences by sampling from
+a profile. It can also generate a "consensus" sequence.
+
+The 
+.B hmmsim
+program is a simulator used for collecting statistics about score
+distributions on random sequences. 
+
+Each program has its own man page.
+
+
+.SH SEE ALSO 
+
+This is a summary man page for the entire HMMER3 package.
+See individual man pages 
+[\fBhmmbuild\fR(1),
+for example] for usage, options, and description of each program in the package.
+
+.PP
+For complete documentation, see the user guide that came with your
+HMMER distribution (Userguide.pdf); or see the HMMER web page
+(http://hmmer.org/).
+
+
+.SH COPYRIGHT
+
+.nf
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi
+
+For additional information on copyright and licensing, see the file
+called COPYRIGHT in your HMMER source distribution, or see the HMMER
+web page 
+(http://hmmer.org/).
+
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi

flax_model/alphafold3/_tools/hmmer/share/man/man1/hmmfetch.1

@@ -0,0 +1,213 @@
+.TH "hmmfetch" 1 "Aug 2023" "HMMER 3.4" "HMMER Manual"
+
+.SH NAME
+hmmfetch \- retrieve profiles from a file
+
+.SH SYNOPSIS
+
+.nf
+\fBhmmfetch\fR [\fIoptions\fR] \fIhmmfile key\fR
+ (retrieve HMM named \fIkey\fR)
+
+\fBhmmfetch \-f \fR[\fIoptions\fR] \fIhmmfile keyfile\fR
+ (retrieve all HMMs listed in \fIkeyfile\fR)
+
+\fBhmmfetch \-\-index \fR[\fIoptions\fR] \fIhmmfile\fR
+ (index \fIhmmfile\fR for fetching)
+.fi
+
+.SH DESCRIPTION
+
+.PP
+Quickly retrieves one or more profile HMMs from an
+.I hmmfile
+(a large Pfam database, for example). 
+
+.PP
+For maximum speed, the 
+.I hmmfile
+should be indexed first, using
+.BR "hmmfetch \-\-index" .
+The index is a binary file named
+.IR hmmfile .ssi.
+However, this is optional, and retrieval will still
+work from unindexed files, albeit much more slowly.
+
+.PP
+The default mode is to retrieve a single profile by name or
+accession, called the
+.IR key .
+For example:
+
+.nf
+    \fB% hmmfetch Pfam-A.hmm Caudal_act\fR
+    \fB% hmmfetch Pfam-A.hmm PF00045\fR
+.fi
+
+.PP
+With the
+.B \-f
+option, a 
+.I keyfile 
+containing a list of one or more keys is read instead. 
+The first whitespace-delimited field on each non-blank non-comment
+line of the
+.I keyfile 
+is used as a 
+.IR key ,
+and any remaining data on the line is ignored. This allows
+a variety of whitespace delimited datafiles to be used
+as a
+.IR keyfile .
+
+.PP
+When using
+.B \-f 
+and a
+.IR keyfile ,
+if 
+.B hmmfile 
+has been indexed, the keys are retrieved in the order
+they occur in the 
+.IR keyfile ,
+but if 
+.B hmmfile 
+isn't indexed, keys are retrieved in the order they occur
+in the 
+.BR hmmfile . 
+This is a side effect of an implementation that allows
+multiple keys to be retrieved even if the
+.B hmmfile 
+is a nonrewindable stream, like a standard input pipe.
+
+.PP 
+In normal use
+(without
+.B \-\-index
+or 
+.B \-f
+options),
+.I hmmfile 
+may be '\-' (dash), which
+means reading input from stdin rather than a file.  
+With the
+.B \-\-index
+option, 
+.I hmmfile
+may not be '\-'; it does not make sense
+to index a standard input stream.
+With the 
+.B \-f 
+option,  
+either 
+.I hmmfile 
+or 
+.I keyfile 
+(but not both) may be '\-'.
+It is often particularly useful to read
+.I keyfile
+from standard input, because this allows
+use to use arbitrary command line invocations to
+create a list of HMM names or accessions, then fetch them all
+to a new file, just with one command.
+
+.PP
+By default, fetched HMMs are printed to standard output in HMMER3 format.
+
+
+.SH OPTIONS
+
+.TP
+.B \-h
+Help; print a brief reminder of command line usage and all available
+options.
+
+.TP
+.B \-f
+The second commandline argument is a 
+.I keyfile
+instead of a single 
+.IR key .
+The first field on each line of the
+.I keyfile 
+is used as a retrieval 
+.I key
+(an HMM name or accession). 
+Blank lines and comment lines (that start with
+a # character) are ignored. 
+
+.TP
+.BI \-o " <f>"
+Output HMM(s) to file
+.I <f>
+instead of to standard output.
+
+.TP
+.B \-O
+Output one retrieved HMM (by
+.IR key )
+to a file named
+.IR key .
+This is a convenience for saving some typing:
+instead of 
+.nf
+ \fB% hmmfetch \-o RRM_1 hmmfile RRM_1\fR
+.fi
+you can just type
+.nf
+ \fB% hmmfetch \-O hmmfile RRM_1\fR
+.fi
+The
+.B \-O 
+option only works if you're retrieving a
+single profile; it is incompatible with 
+.B \-f.
+
+
+
+
+.TP
+.B \-\-index
+Instead of retrieving one or more profiles from
+.IR hmmfile ,
+index the
+.I hmmfile
+for future retrievals.
+This creates a
+.IR hmmfile .ssi
+binary index file.
+
+
+
+.SH SEE ALSO 
+
+See 
+.BR hmmer (1)
+for a master man page with a list of all the individual man pages
+for programs in the HMMER package.
+
+.PP
+For complete documentation, see the user guide that came with your
+HMMER distribution (Userguide.pdf); or see the HMMER web page
+(http://hmmer.org/).
+
+
+
+.SH COPYRIGHT
+
+.nf
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi
+
+For additional information on copyright and licensing, see the file
+called COPYRIGHT in your HMMER source distribution, or see the HMMER
+web page 
+(http://hmmer.org/).
+
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi

flax_model/alphafold3/_tools/hmmer/share/man/man1/hmmlogo.1

@@ -0,0 +1,99 @@
+.TH "hmmlogo" 1 "Aug 2023" "HMMER 3.4" "HMMER Manual"
+
+.SH NAME
+hmmlogo \- produce a conservation logo graphic from a profile
+
+
+.SH SYNOPSIS
+.B hmmlogo
+[\fIoptions\fR]
+.I hmmfile
+
+
+.SH DESCRIPTION
+
+.PP
+.B hmmlogo 
+computes letter height and indel parameters that can be used to 
+produce a profile HMM logo. This tool is essentially a 
+command-line interface for much of the data underlying the Skylign 
+logo server (skylign.org).
+
+By default, 
+.B hmmlogo
+prints out a table of per-position letter heights (dependent on the 
+requested height method), then prints a table of per-position gap
+probabilities. 
+
+In a typical logo, the total height of a stack of letters for one
+position depends on the information content of the position, and 
+that stack height is subdivided according to the emission 
+probabilities of the letters of the alphabet.
+ 
+
+.SH OPTIONS
+
+.TP
+.B \-h
+Help; print a brief reminder of command line usage and all available
+options.
+
+
+.TP
+.B \-\-height_relent_all
+Total height = relative entropy (aka information content); all letters 
+are given a positive height.  (default)
+
+.TP
+.B \-\-height_relent_abovebg
+Total height = relative entropy (aka information content); only letters 
+with above-background probability are given positive height.
+
+.TP
+.B \-\-height_score
+Total height = sums of scores of positive-scoring letters; letter
+height depends on the score of that letter at that position. Only 
+letters with above-background probability (positive score) are 
+given positive height. (Note that only letter height is meaningful -
+stack height has no inherent meaning).
+
+.TP
+.B \-\-no_indel
+Don't print out the indel probability table.
+
+
+.SH SEE ALSO 
+
+See 
+.BR hmmer (1)
+for a master man page with a list of all the individual man pages
+for programs in the HMMER package.
+
+.PP
+For complete documentation, see the user guide that came with your
+HMMER distribution (Userguide.pdf); or see the HMMER web page
+(http://hmmer.org/).
+
+
+
+.SH COPYRIGHT
+
+.nf
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi
+
+For additional information on copyright and licensing, see the file
+called COPYRIGHT in your HMMER source distribution, or see the HMMER
+web page 
+(http://hmmer.org/).
+
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi
+
+
+

flax_model/alphafold3/_tools/hmmer/share/man/man1/hmmpgmd.1

@@ -0,0 +1,219 @@
+.TH "hmmpgmd" 1 "Aug 2023" "HMMER 3.4" "HMMER Manual"
+
+.SH NAME
+hmmpgmd \- daemon for database search web services
+
+
+.SH SYNOPSIS
+.B hmmpgmd
+[\fIoptions\fR]
+
+
+.SH DESCRIPTION
+
+.PP
+The
+.B hmmpgmd 
+program is the daemon that we use internally for the hmmer.org web server. 
+It essentially stands in front of the search programs
+.BR phmmer , 
+.BR hmmsearch ,
+and 
+.BR hmmscan . 
+
+.PP
+To use
+.BR hmmpgmd , 
+first an instance must be started up as a 
+master 
+server, and provided with at least one 
+sequence database
+(using the 
+.B \-\-seqdb
+flag)
+and/or an 
+HMM database
+(using the
+.B \-\-hmmdb
+flag). 
+A sequence database must be in hmmpgmd format, which may be
+produced using 
+.BR esl-reformat .
+An HMM database is of the form produced by 
+.BR hmmbuild .
+The input database(s) will be loaded into memory by the 
+master. When the master has finished loading the database(s), it 
+prints the line:
+"Data loaded into memory. Master is ready."
+
+
+.PP
+After the master is ready, one or more instances of hmmpgmd may
+be started as workers. These workers may be (and typically are) on
+different machines from the master, but must have access to the 
+same database file(s) provided to the master, with the same path. As 
+with the master, each worker loads the database(s) into memory, and 
+indicates completion by printing: "Data loaded into memory. Worker is ready."
+
+
+.PP
+The master process and workers are expected to remain running.
+One or more clients then connect to the master and submit possibly
+many queries. The master distributes the work of a query among the
+workers, collects results, and merges them before responding to the
+client. Two example client programs are included in the HMMER src 
+directory - the C program
+.B hmmc2
+and the perl script
+.BR hmmpgmd_client_example.pl .
+These are intended as examples only, and should be extended as 
+necessary to meet your needs. 
+
+.PP
+A query is submitted to the master from the client as a character
+string. Queries may be the sort that would normally be handled
+by 
+.B phmmer
+(protein sequence vs protein database),
+.B hmmsearch
+(protein HMM query vs protein database), or
+.B hmmscan
+(protein query vs protein HMM database).
+ 
+
+The general form of a client query is to start with a single line
+of the form 
+.BR "@[options]" , 
+followed by multiple lines of text representing either the query HMM 
+or fasta-formatted sequence. The final line of each query is the separator 
+.BR "//" .
+
+
+.PP
+For example, to perform a 
+.B phmmer
+type search of a sequence against a sequence database 
+file, the first line is of the form 
+.BR "@\-\-seqdb 1" ,
+then the fasta-formatted query sequence starting with the header line
+.BR >sequence-name ,
+followed by one or more lines of sequence, and finally the closing
+.BR "//" .
+
+.PP
+To perform an
+.B hmmsearch 
+type search, the query sequence is replaced by the full
+text of a HMMER-format query HMM. 
+
+.PP
+To perform an
+.B hmmscan 
+type search, the text matches that of the 
+.B phmmer
+type search, except that the first line changes to 
+.BR "@\-\-hmmdb 1" .
+
+.PP
+In the hmmpgmd-formatted sequence database file, each sequence
+can be associated with one or more sub-databases. The 
+.B \-\-seqdb
+flag indicates which of these sub-databases will be queried. 
+The HMM database format does not support sub-databases.
+
+
+ 
+
+.SH OPTIONS
+
+.TP
+.B \-h
+Help; print a brief reminder of command line usage and all available
+options.
+
+.TP 
+.BI \-\-master
+Run as the master server.
+
+.TP
+.BI \-\-worker " <s>"
+Run as a worker, connecting to the master server that is running on IP
+address
+.IR <s> .
+
+.TP 
+.BI \-\-cport " <n>"
+Port to use for communication between clients and the master server. 
+The default is 51371.
+
+.TP 
+.BI \-\-wport " <n>"
+Port to use for communication between workers and the master server. 
+The default is 51372.
+
+.TP 
+.BI \-\-ccncts " <n>"
+Maximum number of client connections to accept. The default is 16.
+
+.TP 
+.BI \-\-wcncts " <n>"
+Maximum number of worker connections to accept. The default is 32.
+
+.TP 
+.BI \-\-pid " <f>"
+Name of file into which the process id will be written. 
+
+.TP 
+.BI \-\-seqdb " <f>"
+Name of the file (in
+.B hmmpgmd
+format) containing protein sequences.
+The contents of this file will be cached for searches. 
+
+.TP 
+.BI \-\-hmmdb " <f>"
+Name of the file containing protein HMMs. The contents of this file 
+will be cached for searches.
+
+.TP 
+.BI \-\-cpu " <n>"
+Number of parallel threads to use (for 
+.B \-\-worker
+).
+
+
+.SH SEE ALSO 
+
+See 
+.BR hmmer (1)
+for a master man page with a list of all the individual man pages
+for programs in the HMMER package.
+
+.PP
+For complete documentation, see the user guide that came with your
+HMMER distribution (Userguide.pdf); or see the HMMER web page
+(http://hmmer.org/).
+
+
+
+.SH COPYRIGHT
+
+.nf
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi
+
+For additional information on copyright and licensing, see the file
+called COPYRIGHT in your HMMER source distribution, or see the HMMER
+web page 
+(http://hmmer.org/).
+
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi
+
+
+

flax_model/alphafold3/_tools/hmmer/share/man/man1/hmmpress.1

@@ -0,0 +1,106 @@
+.TH "hmmpress" 1 "Aug 2023" "HMMER 3.4" "HMMER Manual"
+
+.SH NAME
+hmmpress \- prepare a profile database for hmmscan 
+
+.SH SYNOPSIS
+
+.B hmmpress
+[\fIoptions\fR]
+.I hmmfile
+
+
+.SH DESCRIPTION
+
+.PP
+Constructs binary compressed datafiles for 
+.BR hmmscan ,
+starting from a profile database
+.I hmmfile
+in standard HMMER3 format.
+The 
+.B hmmpress
+step is required for
+.B hmmscan
+to work.
+
+.PP
+Four files are created:
+.IB hmmfile .h3m,
+.IB hmmfile .h3i,
+.IB hmmfile .h3f,
+and
+.IB hmmfile .h3p.
+The 
+.IB hmmfile .h3m
+file contains the profile HMMs and their annotation in a binary
+format.
+The 
+.IB hmmfile .h3i
+file is an SSI index for the
+.IB hmmfile .h3m
+file.
+The
+.IB hmmfile .h3f
+file contains precomputed data structures
+for the fast heuristic filter (the MSV filter).
+The
+.IB hmmfile .h3p
+file contains precomputed data structures
+for the rest of each profile.
+
+.PP
+.I hmmfile
+may not be '\-' (dash); running
+.B hmmpress
+on a standard input stream rather than a file
+is not allowed.
+
+
+.SH OPTIONS
+
+.TP
+.B \-h
+Help; print a brief reminder of command line usage and all available
+options.
+
+.TP
+.B \-f
+Force; overwrites any previous hmmpress'ed datafiles. The default is
+to bitch about any existing files and ask you to delete them first.
+
+
+
+
+.SH SEE ALSO 
+
+See 
+.BR hmmer (1)
+for a master man page with a list of all the individual man pages
+for programs in the HMMER package.
+
+.PP
+For complete documentation, see the user guide that came with your
+HMMER distribution (Userguide.pdf); or see the HMMER web page
+(http://hmmer.org/).
+
+
+
+.SH COPYRIGHT
+
+.nf
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi
+
+For additional information on copyright and licensing, see the file
+called COPYRIGHT in your HMMER source distribution, or see the HMMER
+web page 
+(http://hmmer.org/).
+
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi

flax_model/alphafold3/_tools/hmmer/share/man/man1/hmmscan.1

@@ -0,0 +1,470 @@
+.TH "hmmscan" 1 "Aug 2023" "HMMER 3.4" "HMMER Manual"
+
+.SH NAME
+hmmscan \- search sequence(s) against a profile database
+
+
+.SH SYNOPSIS
+.B hmmscan
+[\fIoptions\fR]
+.I hmmdb
+.I seqfile
+
+
+
+.SH DESCRIPTION
+
+.PP
+.B hmmscan 
+is used to search protein sequences against collections 
+of protein profiles. For each sequence in 
+.IR seqfile ,
+use that query sequence to search the target database of
+profiles in
+.IR hmmdb ,
+and output ranked lists of the profiles with the
+most significant matches to the sequence.
+
+.PP
+The 
+.I seqfile 
+may contain more than one query sequence. Each will be searched
+in turn against
+.I hmmdb.
+
+.PP
+The
+.I hmmdb
+needs to be press'ed using 
+.B hmmpress
+before it can be searched with 
+.BR hmmscan . 
+This creates four binary files,
+suffixed
+.BR .h3{fimp} .
+
+.PP 
+The query
+.I seqfile 
+may be '\-' (a dash character), in which case
+the query sequences are read from a
+stdin
+pipe instead of from a file.
+The
+.I hmmdb 
+cannot be read from a
+stdin
+stream, because it needs to have
+those four auxiliary binary files generated by 
+.BR hmmpress .
+
+.PP
+The output format is designed to be human-readable, but is often so
+voluminous that reading it is impractical, and parsing it is a pain. The
+.B \-\-tblout 
+and 
+.B \-\-domtblout 
+options save output in simple tabular formats that are concise and
+easier to parse. 
+The 
+.B \-o
+option allows redirecting the main output, including throwing it away
+in /dev/null.
+
+
+
+.SH OPTIONS
+
+.TP
+.B \-h
+Help; print a brief reminder of command line usage and all available
+options.
+
+
+
+.SH OPTIONS FOR CONTROLLING OUTPUT
+
+.TP 
+.BI \-o " <f>"
+Direct the main human-readable output to a file
+.I <f> 
+instead of the default stdout.
+
+.TP 
+.BI \-\-tblout " <f>"
+Save a simple tabular (space-delimited) file summarizing the
+per-target output, with one data line per homologous target model
+found.
+
+.TP 
+.BI \-\-domtblout " <f>"
+Save a simple tabular (space-delimited) file summarizing the
+per-domain output, with one data line per homologous domain
+detected in a query sequence for each homologous model.
+
+.TP 
+.BI \-\-pfamtblout " <f>"
+Save an especially succinct tabular (space-delimited) file 
+summarizing the per-target output, with one data line per 
+homologous target model found.
+
+
+.TP 
+.B \-\-acc
+Use accessions instead of names in the main output, where available
+for profiles and/or sequences.
+
+.TP 
+.B \-\-noali
+Omit the alignment section from the main output. This can greatly
+reduce the output volume.
+
+.TP 
+.B \-\-notextw
+Unlimit the length of each line in the main output. The default
+is a limit of 120 characters per line, which helps in displaying
+the output cleanly on terminals and in editors, but can truncate
+target profile description lines.
+
+.TP 
+.BI \-\-textw " <n>"
+Set the main output's line length limit to
+.I <n>
+characters per line. The default is 120.
+
+
+
+.SH OPTIONS FOR REPORTING THRESHOLDS
+
+Reporting thresholds control which hits are reported in output files
+(the main output,
+.BR \-\-tblout ,
+and 
+.BR \-\-domtblout ).
+
+.TP
+.BI \-E " <x>"
+In the per-target output, report target profiles with an E-value of <=
+.IR <x> . 
+The default is 10.0, meaning that on average, about 10 false positives
+will be reported per query, so you can see the top of the noise
+and decide for yourself if it's really noise.
+
+.TP
+.BI \-T " <x>"
+Instead of thresholding per-profile output on E-value, instead
+report target profiles with a bit score of >=
+.IR <x> .
+
+.TP
+.BI \-\-domE " <x>"
+In the per-domain output, for target profiles that have already satisfied
+the per-profile reporting threshold, report individual domains
+with a conditional E-value of <=
+.IR <x> . 
+The default is 10.0. 
+A conditional E-value means the expected number of additional false
+positive domains in the smaller search space of those comparisons that
+already satisfied the per-profile reporting threshold (and thus must
+have at least one homologous domain already).
+
+
+.TP
+.BI \-\-domT " <x>"
+Instead of thresholding per-domain output on E-value, instead
+report domains with a bit score of >=
+.IR <x> .
+
+
+
+
+.SH OPTIONS FOR INCLUSION THRESHOLDS
+
+Inclusion thresholds are stricter than reporting thresholds.
+Inclusion thresholds control which hits are considered to be
+reliable enough
+to be included in an output alignment or a subsequent search round.
+In 
+.BR hmmscan ,
+which does not have any alignment output (like 
+.B hmmsearch
+or 
+.BR phmmer ) 
+nor any iterative search steps (like 
+.BR jackhmmer ),
+inclusion thresholds have little effect. They only affect what domains
+get marked as significant (!) or questionable (?) in domain
+output. 
+
+.TP
+.BI \-\-incE " <x>"
+Use an E-value of <=
+.I <x>
+as the per-target inclusion threshold.
+The default is 0.01, meaning that on average, about 1 false positive
+would be expected in every 100 searches with different query
+sequences.
+
+.TP
+.BI \-\-incT " <x>"
+Instead of using E-values for setting the inclusion threshold, instead
+use a bit score of >= 
+.I <x>
+as the per-target inclusion threshold.
+It would be unusual to use bit score thresholds with
+.IR hmmscan ,
+because you don't expect a single score threshold to work for
+different profiles; different profiles have slightly different
+expected score distributions.
+
+.TP
+.BI \-\-incdomE " <x>"
+Use a conditional E-value of <=
+.I <x> 
+as the per-domain inclusion threshold, in targets that have already
+satisfied the overall per-target inclusion threshold.
+The default is 0.01.
+
+.TP
+.BI \-\-incdomT " <x>"
+Instead of using E-values,
+instead use a bit score of >=
+.I <x>
+as the per-domain inclusion threshold.
+As with 
+.B \-\-incT 
+above,
+it would be unusual to use a single bit score threshold in
+.BR hmmscan .
+
+
+
+.SH OPTIONS FOR MODEL-SPECIFIC SCORE THRESHOLDING
+
+Curated profile databases may define specific bit score thresholds for
+each profile, superseding any thresholding based on statistical
+significance alone.
+
+To use these options, the profile must contain the appropriate (GA,
+TC, and/or NC) optional score threshold annotation; this is picked up
+by 
+.B hmmbuild
+from Stockholm format alignment files. Each thresholding option has
+two scores: the per-sequence threshold
+.I <x1>
+and the per-domain threshold
+.IR <x2> .
+These act as if
+.BI \-T " <x1>"
+.BI \-\-incT " <x1>"
+.BI \-\-domT " <x2>"
+.BI \-\-incdomT " <x2>"
+has been applied specifically using each model's curated thresholds.
+
+.TP
+.B \-\-cut_ga
+Use the GA (gathering) bit scores in the model to set
+per-sequence (GA1) and per-domain (GA2) reporting and inclusion
+thresholds. GA thresholds are generally considered to be the
+reliable curated thresholds defining family membership; for example,
+in Pfam, these thresholds define what gets included in Pfam Full
+alignments based on searches with Pfam Seed models.
+
+.TP
+.B \-\-cut_nc
+Use the NC (noise cutoff) bit score thresholds in the model to set
+per-sequence (NC1) and per-domain (NC2) reporting and inclusion
+thresholds. NC thresholds are generally considered to be the score of
+the highest-scoring known false positive.
+
+.TP
+.B \-\-cut_tc
+Use the NC (trusted cutoff) bit score thresholds in the model to set
+per-sequence (TC1) and per-domain (TC2) reporting and inclusion
+thresholds. TC thresholds are generally considered to be the score of
+the lowest-scoring known true positive that is above all known false
+positives. 
+
+
+
+
+.SH CONTROL OF THE ACCELERATION PIPELINE
+
+HMMER3 searches are accelerated in a three-step filter pipeline: the
+MSV filter, the Viterbi filter, and the Forward filter. The first
+filter is the fastest and most approximate; the last is the full
+Forward scoring algorithm. There is also a bias filter step between
+MSV and Viterbi. Targets that pass all the steps in the acceleration
+pipeline are then subjected to postprocessing -- domain
+identification and scoring using the Forward/Backward algorithm.
+
+Changing filter thresholds only removes or includes targets from
+consideration; changing filter thresholds does not alter bit scores,
+E-values, or alignments, all of which are determined solely in
+postprocessing.
+
+.TP
+.B \-\-max
+Turn off all filters, including the bias filter, and run full
+Forward/Backward postprocessing on every target. This increases
+sensitivity somewhat, at a large cost in speed.
+
+.TP
+.BI \-\-F1 " <x>"
+Set the P-value threshold for the MSV filter step.  The default is
+0.02, meaning that roughly 2% of the highest scoring nonhomologous
+targets are expected to pass the filter.
+
+.TP
+.BI \-\-F2 " <x>"
+Set the P-value threshold for the Viterbi filter step.
+The default is 0.001. 
+
+.TP
+.BI \-\-F3 " <x>"
+Set the P-value threshold for the Forward filter step.
+The default is 1e-5.
+
+.TP
+.B \-\-nobias
+Turn off the bias filter. This increases sensitivity somewhat, but can
+come at a high cost in speed, especially if the query has biased
+residue composition (such as a repetitive sequence region, or if it is
+a membrane protein with large regions of hydrophobicity). Without the
+bias filter, too many sequences may pass the filter with biased
+queries, leading to slower than expected performance as the
+computationally intensive Forward/Backward algorithms shoulder an
+abnormally heavy load.
+
+
+
+.SH OTHER OPTIONS
+
+.TP
+.B \-\-nonull2
+Turn off the null2 score corrections for biased composition.
+
+.TP
+.BI \-Z " <x>"
+Assert that the total number of targets in your searches is
+.IR <x> ,
+for the purposes of per-sequence E-value calculations,
+rather than the actual number of targets seen. 
+
+.TP
+.BI \-\-domZ " <x>"
+Assert that the total number of targets in your searches is
+.IR <x> ,
+for the purposes of per-domain conditional E-value calculations,
+rather than the number of targets that passed the reporting thresholds.
+
+.TP
+.BI \-\-seed " <n>"
+Set the random number seed to 
+.IR <n> .
+Some steps in postprocessing require Monte Carlo simulation.  The
+default is to use a fixed seed (42), so that results are exactly
+reproducible. Any other positive integer will give different (but also
+reproducible) results. A choice of 0 uses an arbitrarily chosen seed.
+
+.TP
+.BI \-\-qformat " <s>"
+Assert that input
+.I seqfile
+is in format
+.IR <s> ,
+bypassing format autodetection.
+Common choices for 
+.I <s> 
+include:
+.BR fasta ,
+.BR embl ,
+.BR genbank.
+Alignment formats also work;
+common choices include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+For more information, and for codes for some less common formats,
+see main documentation.
+The string
+.I <s>
+is case-insensitive (\fBfasta\fR or \fBFASTA\fR both work).
+
+
+
+.TP
+.BI \-\-cpu " <n>"
+Set the number of parallel worker threads to 
+.IR <n> .
+The default is 0, meaning off (no thread-level parallelization), because
+.B hmmscan
+is typically i/o bound and the extra overhead of our current
+multithreaded
+implementation isn't worthwhile.
+You can also control this number by setting an environment variable, 
+.IR HMMER_NCPU .
+There is also a master thread, so the actual number of threads that
+HMMER spawns is at least
+.IR <n> +1.
+
+This option is not available if HMMER was compiled with POSIX threads
+support turned off.
+
+
+.TP
+.BI \-\-stall
+For debugging the MPI master/worker version: pause after start, to
+enable the developer to attach debuggers to the running master and
+worker(s) processes. Send SIGCONT signal to release the pause.
+(Under gdb: 
+.BR "(gdb) signal SIGCONT" )
+
+(Only available if optional MPI support was enabled at compile-time.)
+
+.TP
+.B \-\-mpi
+Run under MPI control with master/worker parallelization (using
+.BR mpirun ,
+for example, or equivalent). Only available if optional MPI support
+was enabled at compile-time.
+
+
+
+
+.SH SEE ALSO 
+
+See 
+.BR hmmer (1)
+for a master man page with a list of all the individual man pages
+for programs in the HMMER package.
+
+.PP
+For complete documentation, see the user guide that came with your
+HMMER distribution (Userguide.pdf); or see the HMMER web page
+(http://hmmer.org/).
+
+
+
+.SH COPYRIGHT
+
+.nf
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi
+
+For additional information on copyright and licensing, see the file
+called COPYRIGHT in your HMMER source distribution, or see the HMMER
+web page 
+(http://hmmer.org/).
+
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi
+

flax_model/alphafold3/_tools/hmmer/share/man/man1/hmmsearch.1

@@ -0,0 +1,444 @@
+.TH "hmmsearch" 1 "Aug 2023" "HMMER 3.4" "HMMER Manual"
+
+.SH NAME
+hmmsearch \- search profile(s) against a sequence database
+
+
+.SH SYNOPSIS
+.B hmmsearch
+[\fIoptions\fR]
+.I hmmfile
+.I seqdb
+
+
+.SH DESCRIPTION
+
+.PP
+.B hmmsearch 
+is used to search one or more profiles against a sequence database.
+For each profile in 
+.IR hmmfile ,
+use that query profile to search the target database of sequences in
+.IR seqdb ,
+and output ranked lists of the sequences with the most significant
+matches to the profile.
+To build profiles from multiple alignments, see
+.BR hmmbuild .
+
+.PP 
+Either the query
+.I hmmfile
+or the target
+.I seqdb 
+may be '\-' (a dash character), in which case
+the query profile or target database input will be read from a
+stdin
+pipe instead of from a
+file. Only one input source can come through
+stdin,
+not both.
+An exception is that if the
+.I hmmfile 
+contains more than one profile query, then
+.I seqdb 
+cannot come from stdin, because we can't rewind the
+streaming target database to search it with another profile. 
+
+.PP
+The output format is designed to be human-readable, but is often so
+voluminous that reading it is impractical, and parsing it is a pain. The
+.B \-\-tblout 
+and 
+.B \-\-domtblout 
+options save output in simple tabular formats that are concise and
+easier to parse.
+The 
+.B \-o
+option allows redirecting the main output, including throwing it away
+in /dev/null.
+
+
+
+.SH OPTIONS
+
+.TP
+.B \-h
+Help; print a brief reminder of command line usage and all available
+options.
+
+
+
+.SH OPTIONS FOR CONTROLLING OUTPUT
+
+.TP 
+.BI \-o " <f>"
+Direct the main human-readable output to a file
+.I <f> 
+instead of the default stdout.
+
+.TP
+.BI \-A " <f>"
+Save a multiple alignment of all significant hits (those satisfying
+.IR "inclusion thresholds" )
+to the file 
+.IR <f> .
+
+.TP 
+.BI \-\-tblout " <f>"
+Save a simple tabular (space-delimited) file summarizing the
+per-target output, with one data line per homologous target sequence
+found.
+
+.TP 
+.BI \-\-domtblout " <f>"
+Save a simple tabular (space-delimited) file summarizing the
+per-domain output, with one data line per homologous domain
+detected in a query sequence for each homologous model.
+
+.TP 
+.B \-\-acc
+Use accessions instead of names in the main output, where available
+for profiles and/or sequences.
+
+.TP 
+.B \-\-noali
+Omit the alignment section from the main output. This can greatly
+reduce the output volume.
+
+.TP 
+.B \-\-notextw
+Unlimit the length of each line in the main output. The default
+is a limit of 120 characters per line, which helps in displaying
+the output cleanly on terminals and in editors, but can truncate
+target profile description lines.
+
+.TP 
+.BI \-\-textw " <n>"
+Set the main output's line length limit to
+.I <n>
+characters per line. The default is 120.
+
+
+
+.SH OPTIONS CONTROLLING REPORTING THRESHOLDS
+
+Reporting thresholds control which hits are reported in output files
+(the main output,
+.BR \-\-tblout ,
+and 
+.BR \-\-domtblout ).
+Sequence hits and domain hits are ranked by statistical significance
+(E-value) and output is generated in two sections called per-target
+and per-domain output. In per-target output, by default, all
+sequence hits with an E-value <= 10 are reported. In the per-domain
+output, for each target that has passed per-target reporting
+thresholds, all domains satisfying per-domain reporting thresholds are
+reported. By default, these are domains with conditional E-values of
+<= 10. The following options allow you to change the default
+E-value reporting thresholds, or to use bit score thresholds instead.
+
+
+.TP
+.BI \-E " <x>"
+In the per-target output, report target sequences with an E-value of <=
+.IR <x> . 
+The default is 10.0, meaning that on average, about 10 false positives
+will be reported per query, so you can see the top of the noise
+and decide for yourself if it's really noise.
+
+.TP
+.BI \-T " <x>"
+Instead of thresholding per-profile output on E-value, instead
+report target sequences with a bit score of >=
+.IR <x> .
+
+.TP
+.BI \-\-domE " <x>"
+In the per-domain output, for target sequences that have already satisfied
+the per-profile reporting threshold, report individual domains
+with a conditional E-value of <=
+.IR <x> . 
+The default is 10.0. 
+A conditional E-value means the expected number of additional false
+positive domains in the smaller search space of those comparisons that
+already satisfied the per-target reporting threshold (and thus must
+have at least one homologous domain already).
+
+
+.TP
+.BI \-\-domT " <x>"
+Instead of thresholding per-domain output on E-value, instead
+report domains with a bit score of >=
+.IR <x> .
+
+
+
+
+.SH OPTIONS FOR INCLUSION THRESHOLDS
+
+Inclusion thresholds are stricter than reporting thresholds.
+Inclusion thresholds control which hits are considered to be reliable
+enough to be included in an output alignment or a subsequent search
+round, or marked as significant ("!") as opposed to questionable ("?")
+in domain output.
+
+.TP
+.BI \-\-incE " <x>"
+Use an E-value of <=
+.I <x>
+as the per-target inclusion threshold.
+The default is 0.01, meaning that on average, about 1 false positive
+would be expected in every 100 searches with different query
+sequences.
+
+.TP
+.BI \-\-incT " <x>"
+Instead of using E-values for setting the inclusion threshold, instead
+use a bit score of >= 
+.I <x>
+as the per-target inclusion threshold.
+By default this option is unset.
+
+.TP
+.BI \-\-incdomE " <x>"
+Use a conditional E-value of <=
+.I <x> 
+as the per-domain inclusion threshold, in targets that have already
+satisfied the overall per-target inclusion threshold.
+The default is 0.01.
+
+.TP
+.BI \-\-incdomT " <x>"
+Instead of using E-values,
+use a bit score of >=
+.I <x>
+as the per-domain inclusion threshold.
+
+
+
+.SH OPTIONS FOR MODEL-SPECIFIC SCORE THRESHOLDING
+
+Curated profile databases may define specific bit score thresholds for
+each profile, superseding any thresholding based on statistical
+significance alone.
+
+To use these options, the profile must contain the appropriate (GA,
+TC, and/or NC) optional score threshold annotation; this is picked up
+by 
+.B hmmbuild
+from Stockholm format alignment files. Each thresholding option has
+two scores: the per-sequence threshold <x1> and the per-domain
+threshold <x2>
+These act as if
+.BI \-T " <x1>"
+.BI \-\-incT " <x1>"
+.BI \-\-domT " <x2>"
+.BI \-\-incdomT " <x2>"
+has been applied specifically using each model's curated thresholds.
+
+.TP
+.B \-\-cut_ga
+Use the GA (gathering) bit scores in the model to set
+per-sequence (GA1) and per-domain (GA2) reporting and inclusion
+thresholds. GA thresholds are generally considered to be the
+reliable curated thresholds defining family membership; for example,
+in Pfam, these thresholds define what gets included in Pfam Full
+alignments based on searches with Pfam Seed models.
+
+.TP
+.B \-\-cut_nc
+Use the NC (noise cutoff) bit score thresholds in the model to set
+per-sequence (NC1) and per-domain (NC2) reporting and inclusion
+thresholds. NC thresholds are generally considered to be the score of
+the highest-scoring known false positive.
+
+.TP
+.B \-\-cut_tc
+Use the TC (trusted cutoff) bit score thresholds in the model to set
+per-sequence (TC1) and per-domain (TC2) reporting and inclusion
+thresholds. TC thresholds are generally considered to be the score of
+the lowest-scoring known true positive that is above all known false
+positives. 
+
+
+
+
+.SH OPTIONS CONTROLLING THE ACCELERATION PIPELINE
+
+HMMER3 searches are accelerated in a three-step filter pipeline: the
+MSV filter, the Viterbi filter, and the Forward filter. The first
+filter is the fastest and most approximate; the last is the full
+Forward scoring algorithm. There is also a bias filter step between
+MSV and Viterbi. Targets that pass all the steps in the acceleration
+pipeline are then subjected to postprocessing -- domain
+identification and scoring using the Forward/Backward algorithm.
+
+Changing filter thresholds only removes or includes targets from
+consideration; changing filter thresholds does not alter bit scores,
+E-values, or alignments, all of which are determined solely in
+postprocessing.
+
+.TP
+.B \-\-max
+Turn off all filters, including the bias filter, and run full
+Forward/Backward postprocessing on every target. This increases
+sensitivity somewhat, at a large cost in speed.
+
+.TP
+.BI \-\-F1 " <x>"
+Set the P-value threshold for the MSV filter step.  The default is
+0.02, meaning that roughly 2% of the highest scoring nonhomologous
+targets are expected to pass the filter.
+
+.TP
+.BI \-\-F2 " <x>"
+Set the P-value threshold for the Viterbi filter step.
+The default is 0.001. 
+
+.TP
+.BI \-\-F3 " <x>"
+Set the P-value threshold for the Forward filter step.
+The default is 1e-5.
+
+.TP
+.B \-\-nobias
+Turn off the bias filter. This increases sensitivity somewhat, but can
+come at a high cost in speed, especially if the query has biased
+residue composition (such as a repetitive sequence region, or if it is
+a membrane protein with large regions of hydrophobicity). Without the
+bias filter, too many sequences may pass the filter with biased
+queries, leading to slower than expected performance as the
+computationally intensive Forward/Backward algorithms shoulder an
+abnormally heavy load.
+
+
+
+.SH OTHER OPTIONS
+
+.TP
+.B \-\-nonull2
+Turn off the null2 score corrections for biased composition.
+
+.TP
+.BI \-Z " <x>"
+Assert that the total number of targets in your searches is
+.IR <x> ,
+for the purposes of per-sequence E-value calculations,
+rather than the actual number of targets seen. 
+
+.TP
+.BI \-\-domZ " <x>"
+Assert that the total number of targets in your searches is
+.IR <x> ,
+for the purposes of per-domain conditional E-value calculations,
+rather than the number of targets that passed the reporting thresholds.
+
+.TP
+.BI \-\-seed " <n>"
+Set the random number seed to 
+.IR <n> .
+Some steps in postprocessing require Monte Carlo simulation.  The
+default is to use a fixed seed (42), so that results are exactly
+reproducible. Any other positive integer will give different (but also
+reproducible) results. A choice of 0 uses a randomly chosen seed.
+
+.TP
+.BI \-\-tformat " <s>"
+Assert that target sequence file
+.I seqfile
+is in format
+.IR <s> ,
+bypassing format autodetection.
+Common choices for 
+.I <s> 
+include:
+.BR fasta ,
+.BR embl ,
+.BR genbank.
+Alignment formats also work;
+common choices include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+For more information, and for codes for some less common formats,
+see main documentation.
+The string
+.I <s>
+is case-insensitive (\fBfasta\fR or \fBFASTA\fR both work).
+
+.TP
+.BI \-\-cpu " <n>"
+Set the number of parallel worker threads to 
+.IR <n> .
+On multicore machines, the default is 2.
+You can also control this number by setting an environment variable, 
+.IR HMMER_NCPU .
+There is also a master thread, so the actual number of threads that
+HMMER spawns is
+.IR <n> +1.
+
+This option is not available if HMMER was compiled with POSIX threads
+support turned off.
+
+
+.TP
+.BI \-\-stall
+For debugging the MPI master/worker version: pause after start, to
+enable the developer to attach debuggers to the running master and
+worker(s) processes. Send SIGCONT signal to release the pause.
+(Under gdb: 
+.BR "(gdb) signal SIGCONT" )
+(Only available if optional MPI support was enabled at compile-time.)
+
+
+
+.TP
+.B \-\-mpi
+Run under MPI control with master/worker parallelization (using
+.BR mpirun ,
+for example, or equivalent). Only available if optional MPI support
+was enabled at compile-time.
+
+
+
+
+
+
+.SH SEE ALSO 
+
+See 
+.BR hmmer (1)
+for a master man page with a list of all the individual man pages
+for programs in the HMMER package.
+
+.PP
+For complete documentation, see the user guide that came with your
+HMMER distribution (Userguide.pdf); or see the HMMER web page
+(http://hmmer.org/).
+
+
+
+.SH COPYRIGHT
+
+.nf
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi
+
+For additional information on copyright and licensing, see the file
+called COPYRIGHT in your HMMER source distribution, or see the HMMER
+web page 
+(http://hmmer.org/).
+
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi
+
+
+

flax_model/alphafold3/_tools/hmmer/share/man/man1/hmmsim.1

@@ -0,0 +1,515 @@
+.TH "hmmsim" 1 "Aug 2023" "HMMER 3.4" "HMMER Manual"
+
+.SH NAME
+hmmsim \- collect profile score distributions on random sequences
+
+.SH SYNOPSIS
+.B hmmsim
+[\fIoptions\fR]
+.I hmmfile
+
+
+.SH DESCRIPTION
+
+.PP
+The 
+.B hmmsim 
+program generates random sequences, scores them with the model(s) in 
+.IR hmmfile ,
+and outputs various sorts of histograms, plots, and fitted
+distributions for the resulting scores.
+
+.PP
+.B hmmsim
+is not a mainstream part of the HMMER package and most users would have
+no reason to use it. It is used to develop and test the statistical
+methods used to determine P-values and E-values in HMMER3. For
+example, it was used to generate most of the results in a 2008 paper
+on H3's local alignment statistics (PLoS Comp Bio 4:e1000069, 2008;
+http://www.ploscompbiol.org/doi/pcbi.1000069). 
+
+.PP
+Because it is a research testbed, you should not expect it to be as
+robust as other programs in the package. For example, options may
+interact in weird ways; we haven't tested nor tried to anticipate all
+different possible combinations.
+
+.PP
+The main task is to fit a maximum likelihood Gumbel distribution to
+Viterbi scores or an maximum likelihood exponential tail to
+high-scoring Forward scores, and to test that these fitted
+distributions obey the conjecture that lambda ~ log_2 for both the
+Viterbi Gumbel and the Forward exponential tail. 
+
+.PP
+The output is a table of numbers, one row for each model. Four
+different parametric fits to the score data are tested: (1) maximum
+likelihood fits to both location (mu/tau) and slope (lambda)
+parameters; (2) assuming lambda=log_2, maximum likelihood fit to the
+location parameter only; (3) same but assuming an edge-corrected
+lambda, using current procedures in H3 [Eddy, 2008]; and (4) using
+both parameters determined by H3's current procedures. The standard
+simple, quick and dirty statistic for goodness-of-fit is 'E@10', the
+calculated E-value of the 10th ranked top hit, which we expect to be
+about 10. 
+
+.PP
+In detail, the columns of the output are:
+
+.TP
+.B name
+Name of the model.
+
+.TP
+.B tailp
+Fraction of the highest scores used to fit the distribution. For
+Viterbi, MSV, and Hybrid scores, this defaults to 1.0 (a Gumbel
+distribution is fitted to all the data). For Forward scores, this
+defaults to 0.02 (an exponential tail is fitted to the highest 2%
+scores).
+
+.TP
+.B mu/tau
+Location parameter for the maximum likelihood fit to the data.
+
+.TP
+.B lambda
+Slope parameter for the maximum likelihood fit to the data.
+
+.TP
+.B E@10
+The E-value calculated for the 10th ranked high score ('E@10') using the ML
+mu/tau and lambda. By definition, this expected to be about 10, if
+E-value estimation were accurate.
+
+.TP
+.B mufix
+Location parameter, for a maximum likelihood fit with a known (fixed)
+slope parameter lambda of log_2 (0.693).
+
+.TP
+.B E@10fix
+The E-value calculated for the 10th ranked score using mufix and the
+expected lambda = log_2 = 0.693.
+
+
+.TP
+.B mufix2
+Location parameter, for a maximum likelihood fit with an
+edge-effect-corrected lambda.
+
+.TP
+.B E@10fix2
+The E-value calculated for the 10th ranked score using mufix2 and the
+edge-effect-corrected lambda.
+
+.TP
+.B pmu
+Location parameter as determined by H3's estimation procedures.
+
+.TP
+.B plambda
+Slope parameter as determined by H3's estimation procedures.
+
+.TP
+.B pE@10
+The E-value calculated for the 10th ranked score using pmu, plambda.
+
+
+.PP
+At the end of this table, one more line is printed, starting with #
+and summarizing the overall CPU time used by the simulations.
+
+.PP
+Some of the optional output files are in xmgrace xy format. xmgrace is
+powerful and freely available graph-plotting software.
+
+
+.SH OPTIONS
+
+.TP
+.B \-h
+Help; print a brief reminder of command line usage and all available
+options.
+
+.TP
+.B \-a
+Collect expected Viterbi alignment length statistics from each
+simulated sequence. This only works with Viterbi scores (the default;
+see
+.BR \-\-vit ). 
+Two additional fields are printed in the output table for
+each model: the mean length of Viterbi alignments, and the standard
+deviation.
+
+.TP
+.B \-v
+(Verbose). Print the scores too, one score per line. 
+
+.TP
+.BI \-L " <n>"
+Set the length of the randomly sampled (nonhomologous) sequences to 
+.IR <n> .
+The default is 100.
+
+
+.TP
+.BI \-N " <n>"
+Set the number of randomly sampled sequences to 
+.IR <n> .
+The default is 1000.
+
+.TP
+.B \-\-mpi
+Run under MPI control with master/worker parallelization (using
+.BR mpirun ,
+for example, or equivalent). Only available if optional MPI support
+was enabled at compile-time.
+
+It is parallelized at the level of sending one profile at a time to an
+MPI worker process, so parallelization only helps if you have more
+than one profile in the 
+.IR hmmfile ,
+and you want to have at least as many profiles as MPI worker
+processes.
+
+
+
+
+.SH OPTIONS CONTROLLING OUTPUT
+
+.TP
+.BI \-o " <f>"
+Save the main output table to a file
+.I <f>
+rather than sending it to stdout.
+
+.TP
+.BI \-\-afile " <f>"
+When collecting Viterbi alignment statistics (the
+.B \-a 
+option), for each sampled sequence, output two fields per
+line to a file
+.IR <f> :
+the length of the optimal alignment, and the Viterbi bit score.
+Requires that the 
+.B \-a
+option is also used. 
+
+.TP
+.BI \-\-efile " <f>"
+Output a rank vs. E-value plot in XMGRACE xy format to file
+.IR <f> .
+The x-axis is the rank of this sequence, from highest score to lowest;
+the y-axis is the E-value calculated for this sequence. E-values are
+calculated using H3's default procedures (i.e. the pmu, plambda
+parameters in the output table). You expect a rough match between rank
+and E-value if E-values are accurately estimated.
+
+
+.TP
+.BI \-\-ffile " <f>"
+Output a "filter power" file to 
+.IR <f> :
+for each model, a line with three fields:
+model name, number of sequences passing the P-value threshold,
+and fraction of sequences passing the P-value threshold. See
+.B \-\-pthresh
+for setting the P-value threshold, which defaults to 0.02 (the default
+MSV filter threshold in H3). The P-values are as determined by H3's
+default procedures (the pmu,plambda parameters in the output table).
+If all is well, you expect to see filter power equal to the predicted
+P-value setting of the threshold.
+
+.TP
+.BI \-\-pfile " <f>"
+Output cumulative survival plots (P(S>x)) to file
+.I <f>
+in XMGRACE xy format. There are three plots:
+(1) the observed score distribution; 
+(2) the maximum likelihood fitted distribution;
+(3) a maximum likelihood fit to the location parameter (mu/tau) while
+    assuming lambda=log_2.
+ 
+.TP
+.BI \-\-xfile " <f>"
+Output the bit scores as a binary array of double-precision floats (8
+bytes per score) to file
+.IR <f> .
+Programs like Easel's 
+.B esl-histplot
+can read such binary files. This is useful when generating extremely
+large sample sizes.
+
+
+.SH OPTIONS CONTROLLING MODEL CONFIGURATION (MODE)
+
+H3 only uses multihit local alignment (
+.B \-\-fs 
+mode), and this is where we believe the statistical fits. 
+Unihit local alignment scores (Smith/Waterman; 
+.B \-\-sw
+mode) also obey our statistical conjectures.
+Glocal alignment statistics (either multihit or unihit) are
+still not adequately understood nor adequately fitted.
+
+.TP
+.B \-\-fs
+Collect multihit local alignment scores. This is the default.
+"fs" comes from HMMER2's historical terminology for multihit local
+alignment as 'fragment search mode'.
+
+.TP
+.B \-\-sw
+Collect unihit local alignment scores. The H3 J state is disabled.
+"sw" comes from HMMER2's historical terminology for unihit local
+alignment as 'Smith/Waterman search mode'.
+
+.TP
+.B \-\-ls
+Collect multihit glocal alignment scores. In glocal (global/local)
+alignment, the entire model must align, to a subsequence of the
+target. The H3 local entry/exit transition probabilities are
+disabled. 'ls' comes from HMMER2's historical terminology for multihit local
+alignment as 'local search mode'.
+
+.TP
+.B \-\-s
+Collect unihit glocal alignment scores.  Both the H3 J state and local
+entry/exit transition probabilities are disabled. 's' comes from
+HMMER2's historical terminology for unihit glocal alignment.
+
+
+
+.SH OPTIONS CONTROLLING SCORING ALGORITHM
+
+.TP
+.B \-\-vit
+Collect Viterbi maximum likelihood alignment scores. This is the default.
+
+.TP
+.B \-\-fwd
+Collect Forward log-odds likelihood scores, summed over alignment ensemble.
+
+.TP
+.B \-\-hyb
+Collect 'Hybrid' scores, as described in papers by Yu and Hwa (for
+instance, Bioinformatics 18:864, 2002). These involve calculating a
+Forward matrix and taking the maximum cell value. The number itself is
+statistically somewhat unmotivated, but the distribution is expected
+be a well-behaved extreme value distribution (Gumbel).
+
+.TP
+.B \-\-msv
+Collect MSV (multiple ungapped segment Viterbi) scores, using H3's
+main acceleration heuristic.
+
+.TP
+.B \-\-fast
+For any of the above options, use H3's optimized production
+implementation (using SIMD vectorization). The default is to use the
+"generic" implementation (slow and non-vectorized). The optimized
+implementations sacrifice a small amount of numerical precision. This
+can introduce confounding noise into statistical simulations and fits,
+so when one gets super-concerned about exact details, it's better to
+be able to factor that source of noise out.
+
+.SH OPTIONS CONTROLLING FITTED TAIL MASSES FOR FORWARD 
+
+In some experiments, it was useful to fit Forward scores to a range of
+different tail masses, rather than just one. These options provide a
+mechanism for fitting an evenly-spaced range of different tail masses.
+For each different tail mass, a line is generated in the output.
+
+.TP
+.BI \-\-tmin " <x>"
+Set the lower bound on the tail mass distribution. (The default is
+0.02 for the default single tail mass.)
+
+.TP
+.BI \-\-tmax " <x>"
+Set the upper bound on the tail mass distribution. (The default is
+0.02 for the default single tail mass.)
+
+.TP
+.BI \-\-tpoints " <n>"
+Set the number of tail masses to sample, starting from
+.B \-\-tmin
+and ending at 
+.BR \-\-tmax .
+(The default is 1, for the default 0.02 single tail mass.)
+
+.TP
+.B \-\-tlinear
+Sample a range of tail masses with uniform linear spacing. The default
+is to use uniform logarithmic spacing.
+
+
+
+.SH OPTIONS CONTROLLING H3 PARAMETER ESTIMATION METHODS
+
+H3 uses three short random sequence simulations to estimating the
+location parameters for the expected score distributions for MSV
+scores, Viterbi scores, and Forward scores. These options allow these
+simulations to be modified.
+
+.TP
+.BI \-\-EmL " <n>"
+Sets the sequence length in simulation that estimates the location
+parameter mu for MSV E-values. Default is 200.
+
+.TP
+.BI \-\-EmN " <n>"
+Sets the number of sequences in simulation that estimates the location
+parameter mu for MSV E-values. Default is 200.
+
+.TP
+.BI \-\-EvL " <n>"
+Sets the sequence length in simulation that estimates the location
+parameter mu for Viterbi E-values. Default is 200.
+
+.TP
+.BI \-\-EvN " <n>"
+Sets the number of sequences in simulation that estimates the location
+parameter mu for Viterbi E-values. Default is 200.
+
+.TP
+.BI \-\-EfL " <n>"
+Sets the sequence length in simulation that estimates the location
+parameter tau for Forward E-values. Default is 100.
+
+.TP
+.BI \-\-EfN " <n>"
+Sets the number of sequences in simulation that estimates the location
+parameter tau for Forward E-values. Default is 200.
+
+.TP
+.BI \-\-Eft " <x>"
+Sets the tail mass fraction to fit in the simulation that estimates
+the location parameter tau for Forward evalues. Default is 0.04.
+
+
+.SH DEBUGGING OPTIONS
+
+.TP
+.B \-\-stall
+For debugging the MPI master/worker version: pause after start, to
+enable the developer to attach debuggers to the running master and
+worker(s) processes. Send SIGCONT signal to release the pause.
+(Under gdb: 
+.IR "(gdb) signal SIGCONT" )
+(Only available if optional MPI support was enabled at compile-time.)
+
+.TP
+.BI \-\-seed " <n>"
+Set the random number seed to 
+.IR <n> .
+The default is 0, which makes the random number generator use
+an arbitrary seed, so that different runs of 
+.B hmmsim 
+will almost certainly generate a different statistical sample.
+For debugging, it is useful to force reproducible results, by
+fixing a random number seed.
+
+
+
+.SH EXPERIMENTAL OPTIONS
+
+These options were used in a small variety of different exploratory
+experiments.
+
+.TP
+.B \-\-bgflat 
+Set the background residue distribution to a uniform distribution,
+both for purposes of the null model used in calculating scores, and
+for generating the random sequences. The default is to use a standard
+amino acid background frequency distribution.
+
+.TP
+.B \-\-bgcomp
+Set the background residue distribution to the mean composition of the
+profile. This was used in exploring some of the effects of biased
+composition.
+
+.TP
+.B \-\-x\-no\-lengthmodel
+Turn the H3 target sequence length model off. Set the self-transitions
+for N,C,J and the null model to 350/351 instead; this emulates HMMER2.
+Not a good idea in general. This was used to demonstrate one of the
+main H2 vs. H3 differences.
+
+.TP
+.BI \-\-nu " <x>"
+Set the nu parameter for the MSV algorithm -- the expected number of
+ungapped local alignments per target sequence. The default is 2.0,
+corresponding to a E->J transition probability of 0.5. This was used
+to test whether varying nu has significant effect on result (it
+doesn't seem to, within reason).
+This option
+only works if
+.B \-\-msv
+is selected (it only affects MSV),
+and it will not work with 
+.B \-\-fast
+(because the optimized implementations are hardwired to assume nu=2.0).
+
+.TP
+.BI \-\-pthresh " <x>"
+Set the filter P-value threshold to use in generating filter power
+files with
+.BR \-\-ffile .
+The default is 0.02 (which would be appropriate for testing MSV
+scores, since this is the default MSV filter threshold in H3's
+acceleration pipeline.) Other appropriate choices (matching defaults
+in the acceleration pipeline) would be 0.001 for
+Viterbi, and 1e-5 for Forward.
+
+
+
+
+
+.SH SEE ALSO 
+
+See 
+.BR hmmer (1)
+for a master man page with a list of all the individual man pages
+for programs in the HMMER package.
+
+.PP
+For complete documentation, see the user guide that came with your
+HMMER distribution (Userguide.pdf); or see the HMMER web page
+(http://hmmer.org/).
+
+
+
+.SH COPYRIGHT
+
+.nf
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi
+
+For additional information on copyright and licensing, see the file
+called COPYRIGHT in your HMMER source distribution, or see the HMMER
+web page 
+(http://hmmer.org/).
+
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi
+
+
+
+
+
+
+
+
+
+
+
+
+

flax_model/alphafold3/_tools/hmmer/share/man/man1/jackhmmer.1

@@ -0,0 +1,808 @@
+.TH "jackhmmer" 1 "Aug 2023" "HMMER 3.4" "HMMER Manual"
+
+.SH NAME
+jackhmmer \- iteratively search sequence(s) against a sequence database
+
+.SH SYNOPSIS
+.B jackhmmer
+[\fIoptions\fR]
+.I seqfile
+.I seqdb
+
+.SH DESCRIPTION
+
+.PP
+.B jackhmmer
+iteratively searches each query sequence in 
+.I seqfile
+against the target sequence(s) in
+.IR seqdb .
+The first iteration is identical to a 
+.B phmmer
+search.
+For the next iteration,
+a multiple alignment of the query together with all target sequences
+satisfying 
+inclusion thresholds
+is assembled, a profile is constructed from this alignment
+(identical to using
+.B hmmbuild
+on the alignment), and profile search of the
+.I seqdb
+is done (identical to an
+.B hmmsearch
+with the profile).
+
+
+.PP 
+The query
+.I seqfile 
+may be '\-' (a dash character), in which case
+the query sequences are read from a stdin pipe instead of from a
+file.
+
+The
+.I seqdb 
+needs to be a 'normal' sequence file. It cannot be read from a stdin stream, because
+.B jackhmmer
+needs to do multiple passes over the database. It cannot be a
+compressed (gzipped) file either, because we treat gzipped files
+essentially as stdin streams, calling an external decompression
+program. 
+
+
+.PP
+The output format is designed to be human-readable, but is often so
+voluminous that reading it is impractical, and parsing it is a pain. The
+.B \-\-tblout 
+and 
+.B \-\-domtblout 
+options save output in simple tabular formats that are concise and
+easier to parse.
+The 
+.B \-o
+option allows redirecting the main output, including throwing it away
+in /dev/null.
+
+
+.SH OPTIONS
+
+.TP
+.B \-h
+Help; print a brief reminder of command line usage and all available
+options.
+
+.TP
+.BI \-N " <n>"
+Set the maximum number of iterations to 
+.IR <n> .
+The default is 5. If N=1, the result is equivalent to a
+.B phmmer
+search.
+
+
+
+
+.SH OPTIONS CONTROLLING OUTPUT
+
+By default, output for each iteration appears on stdout in a somewhat
+human readable, somewhat parseable format. These options allow
+redirecting that output or saving additional kinds of output to files,
+including checkpoint files for each iteration.
+
+.TP 
+.BI \-o " <f>"
+Direct the human-readable output to a file
+.IR <f> .
+
+.TP
+.BI \-A " <f>"
+After the final iteration, save an annotated multiple alignment of all
+hits satisfying inclusion thresholds (also including the original query) to
+.I <f>
+in Stockholm format.
+
+.TP
+.BI \-\-tblout " <f>"
+After the final iteration, save a tabular summary of top sequence hits
+to 
+.I <f>
+in a readily parseable, columnar, whitespace-delimited format.
+
+.TP
+.BI \-\-domtblout " <f>"
+After the final iteration, save a tabular summary of top domain hits
+to 
+.I <f>
+in a readily parseable, columnar, whitespace-delimited format.
+
+.TP
+.BI \-\-chkhmm " prefix"
+At the start of each iteration, checkpoint the query HMM, saving it
+to a file named
+\fIprefix\fR\fB-\fR\fIn\fR\fB.hmm\fR
+where
+.I n
+is the iteration number (from 1..N).
+
+.TP
+.BI \-\-chkali " prefix"
+At the end of each iteration, checkpoint an alignment of all
+domains satisfying inclusion thresholds (e.g. what will become the
+query HMM for the next iteration), 
+saving it
+to a file named
+\fIprefix\fR\fB-\fR\fIn\fR\fB.sto\fR
+in Stockholm format,
+where
+.I n
+is the iteration number (from 1..N).
+
+.TP 
+.B \-\-acc
+Use accessions instead of names in the main output, where available
+for profiles and/or sequences.
+
+.TP 
+.B \-\-noali
+Omit the alignment section from the main output. This can greatly
+reduce the output volume.
+
+.TP 
+.B \-\-notextw
+Unlimit the length of each line in the main output. The default
+is a limit of 120 characters per line, which helps in displaying
+the output cleanly on terminals and in editors, but can truncate
+target profile description lines.
+
+.TP 
+.BI \-\-textw " <n>"
+Set the main output's line length limit to
+.I <n>
+characters per line. The default is 120.
+
+
+
+
+
+
+.SH OPTIONS CONTROLLING SINGLE SEQUENCE SCORING (FIRST ITERATION)
+
+By default, the first iteration uses a search model constructed from a
+single query sequence. This model is constructed using a standard
+20x20 substitution matrix for residue probabilities, and two
+additional parameters for position-independent gap open and gap extend
+probabilities. These options allow the default single-sequence scoring
+parameters to be changed.
+
+.TP
+.BI \-\-popen " <x>"
+Set the gap open probability for a single sequence query model to 
+.IR <x> .
+The default is 0.02. 
+.I <x> 
+must be >= 0 and < 0.5.
+
+.TP
+.BI \-\-pextend " <x>"
+Set the gap extend probability for a single sequence query model to 
+.IR <x> .
+The default is 0.4. 
+.I <x> 
+must be >= 0 and < 1.0.
+
+.TP
+.BI \-\-mx " <s>"
+Obtain residue alignment probabilities from the built-in
+substitution matrix named
+.IR <s> . 
+Several standard matrices are built-in, and do not need to be
+read from files. 
+The matrix name
+.I <s> 
+can be
+PAM30, PAM70, PAM120, PAM240, BLOSUM45, BLOSUM50, BLOSUM62, BLOSUM80,
+or BLOSUM90.
+Only one of the
+.B \-\-mx 
+and
+.B \-\-mxfile
+options may be used.
+
+.TP
+.BI \-\-mxfile " mxfile"
+Obtain residue alignment probabilities from the substitution matrix
+in file
+.IR mxfile .
+The default score matrix is BLOSUM62 (this matrix is internal to
+HMMER and does not have to be available as a file). 
+The format of a substitution matrix
+.I mxfile
+is the standard format accepted by BLAST, FASTA, and other sequence 
+analysis software.
+See
+.B ftp.ncbi.nlm.nih.gov/blast/matrices/
+for example files. (The only
+exception: we require matrices to be square, so for DNA, use files
+like NCBI's NUC.4.4, not NUC.4.2.)
+
+
+.SH OPTIONS CONTROLLING REPORTING THRESHOLDS 
+
+Reporting thresholds control which hits are reported in output files
+(the main output,
+.BR \-\-tblout ,
+and 
+.BR \-\-domtblout ).
+In each iteration, sequence hits and domain hits are ranked by
+statistical significance (E-value) and output is generated in two
+sections called per-target and per-domain output. In per-target
+output, by default, all sequence hits with an E-value <= 10 are
+reported. In the per-domain output, for each target that has passed
+per-target reporting thresholds, all domains satisfying per-domain
+reporting thresholds are reported. By default, these are domains with
+conditional E-values of <= 10. The following options allow you to
+change the default E-value reporting thresholds, or to use bit score
+thresholds instead.
+
+
+.TP
+.BI \-E " <x>"
+Report sequences with E-values <=
+.I <x>
+in per-sequence output. The default is 10.0.
+
+.TP
+.BI \-T " <x>"
+Use a bit score threshold for per-sequence output instead of an
+E-value threshold (any setting of
+.B \-E
+is ignored). Report sequences with a bit score of >=
+.IR <x> .
+By default this option is unset.
+
+.TP 
+.BI \-Z " <x>"
+Declare the total size of the database to be
+.I <x>
+sequences, for purposes of E-value calculation.
+Normally E-values are calculated relative to the size of the database
+you actually searched (e.g. the number of sequences in 
+.IR target_seqdb ).
+In some cases (for instance, if you've split your
+target sequence database into multiple files for parallelization of
+your search), you may know better what the actual size of your search
+space is.
+
+.TP
+.BI \-\-domE " <x>"
+Report domains with conditional E-values <=
+.I <x>
+in per-domain output, in addition to the top-scoring
+domain per significant sequence hit. The default is 10.0.
+
+.TP
+.BI \-\-domT " <x>"
+Use a bit score threshold for per-domain output instead of an
+E-value threshold (any setting of
+.B \-\-domE
+is ignored). Report domains with a bit score of >=
+.I <x>
+in per-domain output, in addition to the top-scoring domain per
+significant sequence hit. By default this option is unset.
+
+.TP 
+.BI \-\-domZ " <x>"
+Declare the number of significant sequences to be
+.I <x>
+sequences, for purposes of conditional E-value calculation for
+additional domain significance.
+Normally conditional E-values are calculated relative to the number of
+sequences passing per-sequence reporting threshold.
+
+
+.SH OPTIONS CONTROLLING INCLUSION THRESHOLDS 
+
+Inclusion thresholds control which hits are included in the multiple
+alignment and profile constructed for the next search iteration.
+By default, 
+a sequence must have a per-sequence
+E-value of <= 0.001 (see
+.B \-E 
+option) to be included, and any additional domains in it besides the
+top-scoring one must have a conditional E-value of <= 0.001 (see 
+.B \-\-domE 
+option). The difference between reporting thresholds and inclusion
+thresholds is that inclusion thresholds control which hits actually
+get used in the next iteration (or the final output multiple alignment
+if the 
+.B \-A 
+option is used), whereas reporting thresholds control what you see in
+output. Reporting thresholds are generally more loose so you can see
+borderline hits in the top of the noise that might be of interest.
+
+.TP
+.BI \-\-incE " <x>"
+Include sequences with E-values <=
+.I <x>
+in subsequent iteration or final
+alignment output by 
+.BR \-A .
+The default is 0.001.
+
+.TP
+.BI \-\-incT " <x>"
+Use a bit score threshold for per-sequence inclusion instead of an
+E-value threshold (any setting of
+.B \-\-incE
+is ignored). Include sequences with a bit score of >=
+.IR <x> .
+By default this option is unset.
+
+.TP
+.BI \-\-incdomE " <x>"
+Include domains with conditional E-values <=
+.I <x>
+in subsequent iteration or final alignment output by
+.BR \-A ,
+in addition to the top-scoring
+domain per significant sequence hit. 
+The default is 0.001.
+
+.TP
+.BI \-\-incdomT " <x>"
+Use a bit score threshold for per-domain inclusion instead of an
+E-value threshold (any setting of
+.B \-\-incdomE
+is ignored). Include domains with a bit score of >=
+.IR <x> .
+By default this option is unset.
+
+
+
+.SH OPTIONS CONTROLLING ACCELERATION HEURISTICS
+
+HMMER3 searches are accelerated in a three-step filter pipeline: the
+MSV filter, the Viterbi filter, and the Forward filter. The first
+filter is the fastest and most approximate; the last is the full
+Forward scoring algorithm, slowest but most accurate. There is also a
+bias filter step between MSV and Viterbi. Targets that pass all the
+steps in the acceleration pipeline are then subjected to
+postprocessing -- domain identification and scoring using the
+Forward/Backward algorithm.
+
+Essentially the only free parameters that control HMMER's heuristic
+filters are the P-value thresholds controlling the expected fraction
+of nonhomologous sequences that pass the filters. Setting the default
+thresholds higher will pass a higher proportion of nonhomologous
+sequence, increasing sensitivity at the expense of speed; conversely,
+setting lower P-value thresholds will pass a smaller proportion,
+decreasing sensitivity and increasing speed. Setting a filter's
+P-value threshold to 1.0 means it will passing all sequences, and
+effectively disables the filter.
+
+Changing filter thresholds only removes or includes targets from
+consideration; changing filter thresholds does not alter bit scores,
+E-values, or alignments, all of which are determined solely in
+postprocessing.
+
+.TP
+.B \-\-max
+Maximum sensitivity.  Turn off all filters, including the bias filter,
+and run full Forward/Backward postprocessing on every target. This
+increases sensitivity slightly, at a large cost in speed.
+
+.TP
+.BI \-\-F1 " <x>"
+First filter threshold; set the P-value threshold for the MSV filter
+step.  The default is 0.02, meaning that roughly 2% of the highest
+scoring nonhomologous targets are expected to pass the filter.
+
+.TP
+.BI \-\-F2 " <x>"
+Second filter threshold; set the P-value threshold for the Viterbi
+filter step.  The default is 0.001.
+
+.TP
+.BI \-\-F3 " <x>"
+Third filter threshold; set the P-value threshold for the Forward
+filter step.  The default is 1e-5.
+
+.TP
+.B \-\-nobias
+Turn off the bias filter. This increases sensitivity somewhat, but can
+come at a high cost in speed, especially if the query has biased
+residue composition (such as a repetitive sequence region, or if it is
+a membrane protein with large regions of hydrophobicity). Without the
+bias filter, too many sequences may pass the filter with biased
+queries, leading to slower than expected performance as the
+computationally intensive Forward/Backward algorithms shoulder an
+abnormally heavy load.
+
+
+
+.SH OPTIONS CONTROLLING PROFILE CONSTRUCTION (LATER ITERATIONS)
+
+.B jackhmmer
+always includes your original query sequence in the alignment result
+at every iteration, and consensus positions are always defined by that
+query sequence. That is, a 
+.B jackhmmer
+profile is always the same length as your original query, at every
+iteration.
+Therefore
+.B jackhmmer
+gives you less control over profile construction than
+.B hmmbuild
+does; it does not have the
+.BR \-\-fast ,
+or
+.BR \-\-hand ,
+or
+.B  \-\-symfrac
+options.
+The only profile construction option available in
+.B jackhmmer
+is
+.BR \-\-fragthresh :
+
+
+.TP
+.BI \-\-fragthresh " <x>"
+We only want to count terminal gaps as deletions if the aligned
+sequence is known to be full-length, not if it is a fragment (for
+instance, because only part of it was sequenced). HMMER uses a simple
+rule to infer fragments: if the sequence length L is less than 
+or equal to a fraction
+.I <x> 
+times the alignment length in columns,
+then the sequence is handled as a fragment. The default is 0.5.
+Setting
+.B \-\-fragthresh 0
+will define no (nonempty) sequence as a fragment; you might want to do
+this if you know you've got a carefully curated alignment of full-length
+sequences.
+Setting
+.B \-\-fragthresh 1
+will define all sequences as fragments; you might want to do this if
+you know your alignment is entirely composed of fragments, such as
+translated short reads in metagenomic shotgun data.
+
+
+
+.SH OPTIONS CONTROLLING RELATIVE WEIGHTS
+
+Whenever a profile is built from a multiple alignment, HMMER uses an
+ad hoc sequence weighting algorithm to downweight closely related
+sequences and upweight distantly related ones. This has the effect of
+making models less biased by uneven phylogenetic representation. For
+example, two identical sequences would typically each receive half the
+weight that one sequence would (and this is why 
+.B jackhmmer 
+isn't concerned about always including your original query sequence in
+each iteration's alignment, even if it finds it again in the database
+you're searching). These options control which algorithm gets used.
+
+.TP
+.B \-\-wpb
+Use the Henikoff position-based sequence weighting scheme [Henikoff
+and Henikoff, J. Mol. Biol. 243:574, 1994].  This is the default.
+
+.TP 
+.B \-\-wgsc 
+Use the Gerstein/Sonnhammer/Chothia weighting algorithm [Gerstein et
+al, J. Mol. Biol. 235:1067, 1994].
+
+.TP 
+.B \-\-wblosum
+Use the same clustering scheme that was used to weight data in
+calculating BLOSUM substitution matrices [Henikoff and Henikoff,
+Proc. Natl. Acad. Sci 89:10915, 1992]. Sequences are single-linkage
+clustered at an identity threshold (default 0.62; see
+.BR \-\-wid )
+and within each cluster of c sequences, each sequence gets relative
+weight 1/c.
+
+.TP
+.B \-\-wnone
+No relative weights. All sequences are assigned uniform weight. 
+
+.TP 
+.BI \-\-wid " <x>"
+Sets the identity threshold used by single-linkage clustering when 
+using 
+.BR \-\-wblosum . 
+Invalid with any other weighting scheme. Default is 0.62.
+
+
+
+
+
+.SH OPTIONS CONTROLLING EFFECTIVE SEQUENCE NUMBER
+
+After relative weights are determined, they are normalized to sum to a
+total effective sequence number, 
+.IR eff_nseq . 
+This number may be the actual number of sequences in the alignment,
+but it is almost always smaller than that.
+The default entropy weighting method 
+(\fB\-\-eent\fR)
+reduces the effective sequence
+number to reduce the information content (relative entropy, or average
+expected score on true homologs) per consensus position. The target
+relative entropy is controlled by a two-parameter function, where the
+two parameters are settable with
+.B \-\-ere
+and 
+.BR \-\-esigma .
+
+.TP
+.B \-\-eent
+Adjust effective sequence number to achieve a specific relative entropy
+per position (see
+.BR \-\-ere ).
+This is the default.
+
+.TP
+.B \-\-eclust
+Set effective sequence number to the number of single-linkage clusters
+at a specific identity threshold (see 
+.BR \-\-eid ).
+This option is not recommended; it's for experiments evaluating
+how much better
+.B \-\-eent
+is.
+
+.TP
+.B \-\-enone
+Turn off effective sequence number determination and just use the
+actual number of sequences. One reason you might want to do this is
+to try to maximize the relative entropy/position of your model, which
+may be useful for short models.
+
+.TP
+.BI \-\-eset " <x>"
+Explicitly set the effective sequence number for all models to 
+.IR <x> .
+
+.TP
+.BI \-\-ere " <x>"
+Set the minimum relative entropy/position target to 
+.IR <x> .
+Requires
+.BR \-\-eent . 
+Default depends on the sequence alphabet; for protein
+sequences, it is 0.59 bits/position.
+
+.TP
+.BI \-\-esigma " <x>"
+Sets the minimum relative entropy contributed by an entire
+model alignment, over its whole length. This has the effect
+of making short models have 
+higher relative entropy per position than 
+.B \-\-ere 
+alone would give. The default is 45.0 bits.
+
+.TP
+.BI \-\-eid " <x>"
+Sets the fractional pairwise identity cutoff used by 
+single linkage clustering with the
+.B \-\-eclust 
+option. The default is 0.62.
+
+
+
+.SH OPTIONS CONTROLLING PRIORS
+
+In profile construction, by default, weighted counts are converted to
+mean posterior probability parameter estimates using mixture Dirichlet
+priors.  Default mixture Dirichlet prior parameters for protein models
+and for nucleic acid (RNA and DNA) models are built in. The following
+options allow you to override the default priors.
+
+.TP
+.B \-\-pnone
+Don't use any priors. Probability parameters will simply be the
+observed frequencies, after relative sequence weighting. 
+
+.TP
+.B \-\-plaplace
+Use a Laplace +1 prior in place of the default mixture Dirichlet
+prior.
+
+
+
+.SH OPTIONS CONTROLLING E-VALUE CALIBRATION
+
+Estimating the location parameters for the expected score
+distributions for MSV filter scores, Viterbi filter scores, and
+Forward scores requires three short random sequence simulations.
+
+.TP
+.BI \-\-EmL " <n>"
+Sets the sequence length in simulation that estimates the location
+parameter mu for MSV filter E-values. Default is 200.
+
+.TP
+.BI \-\-EmN " <n>"
+Sets the number of sequences in simulation that estimates the location
+parameter mu for MSV filter E-values. Default is 200.
+
+.TP
+.BI \-\-EvL " <n>"
+Sets the sequence length in simulation that estimates the location
+parameter mu for Viterbi filter E-values. Default is 200.
+
+.TP
+.BI \-\-EvN " <n>"
+Sets the number of sequences in simulation that estimates the location
+parameter mu for Viterbi filter E-values. Default is 200.
+
+.TP
+.BI \-\-EfL " <n>"
+Sets the sequence length in simulation that estimates the location
+parameter tau for Forward E-values. Default is 100.
+
+.TP
+.BI \-\-EfN " <n>"
+Sets the number of sequences in simulation that estimates the location
+parameter tau for Forward E-values. Default is 200.
+
+.TP
+.BI \-\-Eft " <x>"
+Sets the tail mass fraction to fit in the simulation that estimates
+the location parameter tau for Forward evalues. Default is 0.04.
+
+
+.SH OTHER OPTIONS
+
+.TP
+.B \-\-nonull2
+Turn off the null2 score corrections for biased composition.
+
+.TP
+.BI \-Z " <x>"
+Assert that the total number of targets in your searches is
+.IR <x> ,
+for the purposes of per-sequence E-value calculations,
+rather than the actual number of targets seen. 
+
+.TP
+.BI \-\-domZ " <x>"
+Assert that the total number of targets in your searches is
+.IR <x> ,
+for the purposes of per-domain conditional E-value calculations,
+rather than the number of targets that passed the reporting thresholds.
+
+.TP 
+.BI \-\-seed " <n>"
+Seed the random number generator with
+.IR <n> ,
+an integer >= 0. 
+If 
+.I <n> 
+is >0, any stochastic simulations will be reproducible; the same
+command will give the same results.
+If 
+.I <n>
+is 0, the random number generator is seeded arbitrarily, and
+stochastic simulations will vary from run to run of the same command.
+The default seed is 42.
+
+
+.TP 
+.BI \-\-qformat " <s>"
+Assert that input query
+.I seqfile
+is in format
+.IR <s> ,
+bypassing format autodetection.
+Common choices for 
+.I <s> 
+include:
+.BR fasta ,
+.BR embl ,
+.BR genbank.
+Alignment formats also work;
+common choices include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+.B jackhmmer
+always uses a single sequence query to start its search, so when the input
+.I seqfile
+is an alignment,
+.B jackhmmer
+reads it one unaligned query sequence at a time, not as an alignment.
+For more information, and for codes for some less common formats,
+see main documentation.
+The string
+.I <s>
+is case-insensitive (\fBfasta\fR or \fBFASTA\fR both work).
+
+.TP 
+.BI \-\-tformat " <s>"
+Assert that the input target sequence
+.I seqdb
+is in format 
+.IR <s> .
+See
+.B \-\-qformat
+above for accepted choices for
+.IR <s> .
+
+
+
+.TP
+.BI \-\-cpu " <n>"
+Set the number of parallel worker threads to 
+.IR <n> .
+On multicore machines, the default is 2.
+You can also control this number by setting an environment variable, 
+.IR HMMER_NCPU .
+There is also a master thread, so the actual number of threads that
+HMMER spawns is
+.IR <n> +1.
+
+This option is not available if HMMER was compiled with POSIX threads
+support turned off.
+
+
+
+.TP
+.BI \-\-stall
+For debugging the MPI master/worker version: pause after start, to
+enable the developer to attach debuggers to the running master and
+worker(s) processes. Send SIGCONT signal to release the pause.
+(Under gdb: 
+.BR "(gdb) signal SIGCONT" )
+(Only available if optional MPI support was enabled at compile-time.)
+
+.TP
+.BI \-\-mpi
+Run under MPI control with master/worker parallelization (using
+.BR mpirun ,
+for example, or equivalent). Only available if optional MPI support
+was enabled at compile-time.
+
+
+
+
+
+.SH SEE ALSO 
+
+See 
+.BR hmmer (1)
+for a master man page with a list of all the individual man pages
+for programs in the HMMER package.
+
+.PP
+For complete documentation, see the user guide that came with your
+HMMER distribution (Userguide.pdf); or see the HMMER web page
+(http://hmmer.org/).
+
+
+
+.SH COPYRIGHT
+
+.nf
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi
+
+For additional information on copyright and licensing, see the file
+called COPYRIGHT in your HMMER source distribution, or see the HMMER
+web page 
+(http://hmmer.org/).
+
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi
+
+

flax_model/alphafold3/_tools/hmmer/share/man/man1/phmmer.1

@@ -0,0 +1,523 @@
+.TH "phmmer" 1 "Aug 2023" "HMMER 3.4" "HMMER Manual"
+
+.SH NAME
+phmmer \- search protein sequence(s) against a protein sequence database
+
+
+.SH SYNOPSIS
+.B phmmer
+[\fIoptions\fR]
+.I seqfile
+.I seqdb
+
+
+.SH DESCRIPTION
+
+.PP
+.B phmmer
+is used to search one or more query protein sequences against a protein sequence database.
+For each query sequence in 
+.IR seqfile ,
+use that sequence to search the target database of sequences in
+.IR seqdb ,
+and output ranked lists of the sequences with the most significant
+matches to the query.
+
+.PP 
+Either the query
+.I seqfile
+or the target
+.I seqdb 
+may be '\-' (a dash character), in which case
+the query sequences or target database input will be read from a <stdin> pipe instead of from a
+file. Only one input source can come through <stdin>, not both.
+An exception is that if the
+.I seqfile 
+contains more than one query sequence, then
+.I seqdb 
+cannot come from <stdin>, because we can't rewind the
+streaming target database to search it with another query.
+
+
+.PP
+The output format is designed to be human-readable, but is often so
+voluminous that reading it is impractical, and parsing it is a pain. The
+.B \-\-tblout 
+and 
+.B \-\-domtblout 
+options save output in simple tabular formats that are concise and
+easier to parse.
+The 
+.B \-o
+option allows redirecting the main output, including throwing it away
+in /dev/null.
+
+.SH OPTIONS
+
+.TP
+.B \-h
+Help; print a brief reminder of command line usage and all available
+options.
+
+
+.SH OPTIONS FOR CONTROLLING OUTPUT
+
+.TP 
+.BI \-o " <f>"
+Direct the main human-readable output to a file
+.I <f> 
+instead of the default stdout.
+
+.TP
+.BI \-A " <f>"
+Save a multiple alignment of all significant hits (those satisfying
+inclusion thresholds)
+to the file 
+.I <f>
+in Stockholm format.
+
+.TP 
+.BI \-\-tblout " <f>"
+Save a simple tabular (space-delimited) file summarizing the
+per-target output, with one data line per homologous target sequence
+found.
+
+.TP 
+.BI \-\-domtblout " <f>"
+Save a simple tabular (space-delimited) file summarizing the
+per-domain output, with one data line per homologous domain
+detected in a query sequence for each homologous model.
+
+.TP 
+.B \-\-acc
+Use accessions instead of names in the main output, where available
+for profiles and/or sequences.
+
+.TP 
+.B \-\-noali
+Omit the alignment section from the main output. This can greatly
+reduce the output volume.
+
+.TP 
+.B \-\-notextw
+Unlimit the length of each line in the main output. The default
+is a limit of 120 characters per line, which helps in displaying
+the output cleanly on terminals and in editors, but can truncate
+target profile description lines.
+
+.TP 
+.BI \-\-textw " <n>"
+Set the main output's line length limit to
+.I <n>
+characters per line. The default is 120.
+
+
+
+.SH OPTIONS CONTROLLING SCORING SYSTEM
+
+The probability model in
+.B phmmer
+is constructed by inferring residue probabilities from a standard
+20x20 substitution score matrix, plus two additional parameters for
+position-independent gap open and gap extend probabilities.
+
+.TP
+.BI \-\-popen " <x>"
+Set the gap open probability for a single sequence query model to 
+.IR <x> .
+The default is 0.02. 
+.I <x> 
+must be >= 0 and < 0.5.
+
+.TP
+.BI \-\-pextend " <x>"
+Set the gap extend probability for a single sequence query model to 
+.IR <x> .
+The default is 0.4. 
+.I <x> 
+must be >= 0 and < 1.0.
+
+.TP
+.BI \-\-mx " <s>"
+Obtain residue alignment probabilities from the built-in
+substitution matrix named
+.IR <s> . 
+Several standard matrices are built-in, and do not need to be
+read from files. 
+The matrix name
+.I <s> 
+can be
+PAM30, PAM70, PAM120, PAM240, BLOSUM45, BLOSUM50, BLOSUM62, BLOSUM80,
+or BLOSUM90.
+Only one of the
+.B \-\-mx 
+and
+.B \-\-mxfile
+options may be used.
+
+.TP
+.BI \-\-mxfile " mxfile"
+Obtain residue alignment probabilities from the substitution matrix
+in file
+.IR mxfile .
+The default score matrix is BLOSUM62 (this matrix is internal to
+HMMER and does not have to be available as a file). 
+The format of a substitution matrix
+.I mxfile
+is the standard format accepted by BLAST, FASTA, and other sequence 
+analysis software.
+See ftp.ncbi.nlm.nih.gov/blast/matrices/ for example files. (The only
+exception: we require matrices to be square, so for DNA, use files
+like NCBI's NUC.4.4, not NUC.4.2.)
+
+
+
+.SH OPTIONS CONTROLLING REPORTING THRESHOLDS
+
+Reporting thresholds control which hits are reported in output files
+(the main output,
+.BR \-\-tblout ,
+and 
+.BR \-\-domtblout ).
+Sequence hits and domain hits are ranked by statistical significance
+(E-value) and output is generated in two sections called per-target
+and per-domain output. In per-target output, by default, all
+sequence hits with an E-value <= 10 are reported. In the per-domain
+output, for each target that has passed per-target reporting
+thresholds, all domains satisfying per-domain reporting thresholds are
+reported. By default, these are domains with conditional E-values of
+<= 10. The following options allow you to change the default
+E-value reporting thresholds, or to use bit score thresholds instead.
+
+
+.TP
+.BI \-E " <x>"
+In the per-target output, report target sequences with an E-value of <=
+.IR <x> . 
+The default is 10.0, meaning that on average, about 10 false positives
+will be reported per query, so you can see the top of the noise
+and decide for yourself if it's really noise.
+
+.TP
+.BI \-T " <x>"
+Instead of thresholding per-profile output on E-value, instead
+report target sequences with a bit score of >=
+.IR <x> .
+
+.TP
+.BI \-\-domE " <x>"
+In the per-domain output, for target sequences that have already satisfied
+the per-profile reporting threshold, report individual domains
+with a conditional E-value of <=
+.IR <x> . 
+The default is 10.0. 
+A conditional E-value means the expected number of additional false
+positive domains in the smaller search space of those comparisons that
+already satisfied the per-target reporting threshold (and thus must
+have at least one homologous domain already).
+
+.TP
+.BI \-\-domT " <x>"
+Instead of thresholding per-domain output on E-value, instead
+report domains with a bit score of >=
+.IR <x> .
+
+.SH OPTIONS CONTROLLING INCLUSION THRESHOLDS
+
+Inclusion thresholds are stricter than reporting thresholds. They
+control which hits are included in any output multiple alignment (the
+.B \-A 
+option) and which domains are marked as significant ("!") as opposed
+to questionable ("?")  in domain output.
+
+.TP
+.BI \-\-incE " <x>"
+Use an E-value of <=
+.I <x>
+as the per-target inclusion threshold.
+The default is 0.01, meaning that on average, about 1 false positive
+would be expected in every 100 searches with different query
+sequences.
+
+.TP
+.BI \-\-incT " <x>"
+Instead of using E-values for setting the inclusion threshold, instead
+use a bit score of >= 
+.I <x>
+as the per-target inclusion threshold.
+By default this option is unset.
+
+.TP
+.BI \-\-incdomE " <x>"
+Use a conditional E-value of <=
+.I <x> 
+as the per-domain inclusion threshold, in targets that have already
+satisfied the overall per-target inclusion threshold.
+The default is 0.01.
+
+.TP
+.BI \-\-incdomT " <x>"
+Instead of using E-values,
+use a bit score of >=
+.I <x>
+as the per-domain inclusion threshold.
+By default this option is unset.
+
+
+
+
+.SH OPTIONS CONTROLLING THE ACCELERATION PIPELINE
+
+HMMER3 searches are accelerated in a three-step filter pipeline: the
+MSV filter, the Viterbi filter, and the Forward filter. The first
+filter is the fastest and most approximate; the last is the full
+Forward scoring algorithm, slowest but most accurate. There is also a
+bias filter step between MSV and Viterbi. Targets that pass all the
+steps in the acceleration pipeline are then subjected to
+postprocessing -- domain identification and scoring using the
+Forward/Backward algorithm.
+
+Essentially the only free parameters that control HMMER's heuristic
+filters are the P-value thresholds controlling the expected fraction
+of nonhomologous sequences that pass the filters. Setting the default
+thresholds higher will pass a higher proportion of nonhomologous
+sequence, increasing sensitivity at the expense of speed; conversely,
+setting lower P-value thresholds will pass a smaller proportion,
+decreasing sensitivity and increasing speed. Setting a filter's
+P-value threshold to 1.0 means it will passing all sequences, and
+effectively disables the filter.
+
+Changing filter thresholds only removes or includes targets from
+consideration; changing filter thresholds does not alter bit scores,
+E-values, or alignments, all of which are determined solely in
+postprocessing.
+
+.TP
+.B \-\-max
+Maximum sensitivity.  Turn off all filters, including the bias filter,
+and run full Forward/Backward postprocessing on every target. This
+increases sensitivity slightly, at a large cost in speed.
+
+.TP
+.BI \-\-F1 " <x>"
+First filter threshold; set the P-value threshold for the MSV filter
+step.  The default is 0.02, meaning that roughly 2% of the highest
+scoring nonhomologous targets are expected to pass the filter.
+
+.TP
+.BI \-\-F2 " <x>"
+Second filter threshold; set the P-value threshold for the Viterbi
+filter step.  The default is 0.001.
+
+.TP
+.BI \-\-F3 " <x>"
+Third filter threshold; set the P-value threshold for the Forward
+filter step.  The default is 1e-5.
+
+.TP
+.B \-\-nobias
+Turn off the bias filter. This increases sensitivity somewhat, but can
+come at a high cost in speed, especially if the query has biased
+residue composition (such as a repetitive sequence region, or if it is
+a membrane protein with large regions of hydrophobicity). Without the
+bias filter, too many sequences may pass the filter with biased
+queries, leading to slower than expected performance as the
+computationally intensive Forward/Backward algorithms shoulder an
+abnormally heavy load.
+
+
+
+
+.SH OPTIONS CONTROLLING E-VALUE CALIBRATION
+
+Estimating the location parameters for the expected score
+distributions for MSV filter scores, Viterbi filter scores, and
+Forward scores requires three short random sequence simulations.
+
+.TP
+.BI \-\-EmL " <n>"
+Sets the sequence length in simulation that estimates the location
+parameter mu for MSV filter E-values. Default is 200.
+
+.TP
+.BI \-\-EmN " <n>"
+Sets the number of sequences in simulation that estimates the location
+parameter mu for MSV filter E-values. Default is 200.
+
+.TP
+.BI \-\-EvL " <n>"
+Sets the sequence length in simulation that estimates the location
+parameter mu for Viterbi filter E-values. Default is 200.
+
+.TP
+.BI \-\-EvN " <n>"
+Sets the number of sequences in simulation that estimates the location
+parameter mu for Viterbi filter E-values. Default is 200.
+
+.TP
+.BI \-\-EfL " <n>"
+Sets the sequence length in simulation that estimates the location
+parameter tau for Forward E-values. Default is 100.
+
+.TP
+.BI \-\-EfN " <n>"
+Sets the number of sequences in simulation that estimates the location
+parameter tau for Forward E-values. Default is 200.
+
+.TP
+.BI \-\-Eft " <x>"
+Sets the tail mass fraction to fit in the simulation that estimates
+the location parameter tau for Forward evalues. Default is 0.04.
+
+
+
+
+.SH OTHER OPTIONS
+
+.TP
+.B \-\-nonull2
+Turn off the null2 score corrections for biased composition.
+
+.TP
+.BI \-Z " <x>"
+Assert that the total number of targets in your searches is
+.IR <x> ,
+for the purposes of per-sequence E-value calculations,
+rather than the actual number of targets seen. 
+
+.TP
+.BI \-\-domZ " <x>"
+Assert that the total number of targets in your searches is
+.IR <x> ,
+for the purposes of per-domain conditional E-value calculations,
+rather than the number of targets that passed the reporting thresholds.
+
+.TP 
+.BI \-\-seed " <n>"
+Seed the random number generator with
+.IR <n> ,
+an integer >= 0. 
+If 
+.I <n> 
+is >0, any stochastic simulations will be reproducible; the same
+command will give the same results.
+If 
+.I <n>
+is 0, the random number generator is seeded arbitrarily, and
+stochastic simulations will vary from run to run of the same command.
+The default seed is 42.
+
+.TP 
+.BI \-\-qformat " <s>"
+Assert that input
+.I seqfile
+is in format
+.IR <s> ,
+bypassing format autodetection.
+Common choices for 
+.I <s> 
+include:
+.BR fasta ,
+.BR embl ,
+.BR genbank.
+Alignment formats also work;
+common choices include:
+.BR stockholm , 
+.BR a2m ,
+.BR afa ,
+.BR psiblast ,
+.BR clustal ,
+.BR phylip .
+.B phmmer
+always uses a single sequence query to start its search, so when the input
+.I seqfile
+is an alignment,
+.B phmmer
+reads it one unaligned query sequence at a time, not as an alignment.
+For more information, and for codes for some less common formats,
+see main documentation.
+The string
+.I <s>
+is case-insensitive (\fBfasta\fR or \fBFASTA\fR both work).
+
+.BI \-\-tformat " <s>"
+Assert that target sequence database
+.I seqdb
+is in format
+.IR <s> ,
+bypassing format autodetection.
+See
+.B \-\-qformat
+above for list of accepted format codes for
+.IR <s> .
+
+
+.TP
+.BI \-\-cpu " <n>"
+Set the number of parallel worker threads to 
+.IR <n> .
+On multicore machines, the default is 2.
+You can also control this number by setting an environment variable, 
+.IR HMMER_NCPU .
+There is also a master thread, so the actual number of threads that
+HMMER spawns is
+.IR <n> +1.
+
+This option is not available if HMMER was compiled with POSIX threads
+support turned off.
+
+
+
+.TP
+.BI \-\-stall
+For debugging the MPI master/worker version: pause after start, to
+enable the developer to attach debuggers to the running master and
+worker(s) processes. Send SIGCONT signal to release the pause.
+(Under gdb: 
+.BR "(gdb) signal SIGCONT" )
+(Only available if optional MPI support was enabled at compile-time.)
+
+.TP
+.BI \-\-mpi
+Run under MPI control with master/worker parallelization (using
+.BR mpirun ,
+for example, or equivalent). Only available if optional MPI support
+was enabled at compile-time.
+
+
+
+
+
+.SH SEE ALSO 
+
+See 
+.BR hmmer (1)
+for a master man page with a list of all the individual man pages
+for programs in the HMMER package.
+
+.PP
+For complete documentation, see the user guide that came with your
+HMMER distribution (Userguide.pdf); or see the HMMER web page
+(http://hmmer.org/).
+
+
+
+.SH COPYRIGHT
+
+.nf
+Copyright (C) 2023 Howard Hughes Medical Institute.
+Freely distributed under the BSD open source license.
+.fi
+
+For additional information on copyright and licensing, see the file
+called COPYRIGHT in your HMMER source distribution, or see the HMMER
+web page 
+(http://hmmer.org/).
+
+
+.SH AUTHOR
+
+.nf
+http://eddylab.org
+.fi
+

flax_model/alphafold3/data/cpp/msa_profile_pybind.h

@@ -0,0 +1,24 @@
+/*
+ * Copyright 2024 DeepMind Technologies Limited
+ *
+ * AlphaFold 3 source code is licensed under CC BY-NC-SA 4.0. To view a copy of
+ * this license, visit https://creativecommons.org/licenses/by-nc-sa/4.0/
+ *
+ * To request access to the AlphaFold 3 model parameters, follow the process set
+ * out at https://github.com/google-deepmind/alphafold3. You may only use these
+ * if received directly from Google. Use is subject to terms of use available at
+ * https://github.com/google-deepmind/alphafold3/blob/main/WEIGHTS_TERMS_OF_USE.md
+ */
+
+#ifndef ALPHAFOLD3_SRC_ALPHAFOLD3_DATA_PYTHON_MSA_PROFILE_PYBIND_H_
+#define ALPHAFOLD3_SRC_ALPHAFOLD3_DATA_PYTHON_MSA_PROFILE_PYBIND_H_
+
+#include "pybind11/pybind11.h"
+
+namespace alphafold3 {
+
+void RegisterModuleMsaProfile(pybind11::module m);
+
+}
+
+#endif  // ALPHAFOLD3_SRC_ALPHAFOLD3_DATA_PYTHON_MSA_PROFILE_PYBIND_H_

flax_model/alphafold3/data/featurisation.py

@@ -0,0 +1,96 @@
+
+
+"""AlphaFold 3 featurisation pipeline."""
+
+from collections.abc import Sequence
+import datetime
+import time
+
+from flax_model.alphafold3.common import folding_input
+from flax_model.alphafold3.constants import chemical_components
+from flax_model.alphafold3.model import features
+from flax_model.alphafold3.model.pipeline import pipeline
+import numpy as np
+
+
+def validate_fold_input(fold_input: folding_input.Input):
+  """Validates the fold input contains MSA and templates for featurisation."""
+  for i, chain in enumerate(fold_input.protein_chains):
+    if chain.unpaired_msa is None:
+      raise ValueError(f'Protein chain {i + 1} is missing unpaired MSA.')
+    if chain.paired_msa is None:
+      raise ValueError(f'Protein chain {i + 1} is missing paired MSA.')
+    if chain.templates is None:
+      raise ValueError(f'Protein chain {i + 1} is missing Templates.')
+  for i, chain in enumerate(fold_input.rna_chains):
+    if chain.unpaired_msa is None:
+      raise ValueError(f'RNA chain {i + 1} is missing unpaired MSA.')
+
+
+def featurise_input(
+    fold_input: folding_input.Input,
+    ccd: chemical_components.Ccd,
+    buckets: Sequence[int] | None,
+    ref_max_modified_date: datetime.date | None = None,
+    conformer_max_iterations: int | None = None,
+    resolve_msa_overlaps: bool = True,
+    verbose: bool = False,
+) -> Sequence[features.BatchDict]:
+  """Featurise the folding input.
+
+  Args:
+    fold_input: The input to featurise.
+    ccd: The chemical components dictionary.
+    buckets: Bucket sizes to pad the data to, to avoid excessive re-compilation
+      of the model. If None, calculate the appropriate bucket size from the
+      number of tokens. If not None, must be a sequence of at least one integer,
+      in strictly increasing order. Will raise an error if the number of tokens
+      is more than the largest bucket size.
+    ref_max_modified_date: Optional maximum date that controls whether to allow
+      use of model coordinates for a chemical component from the CCD if RDKit
+      conformer generation fails and the component does not have ideal
+      coordinates set. Only for components that have been released before this
+      date the model coordinates can be used as a fallback.
+    conformer_max_iterations: Optional override for maximum number of iterations
+      to run for RDKit conformer search.
+    resolve_msa_overlaps: Whether to deduplicate unpaired MSA against paired
+      MSA. The default behaviour matches the method described in the AlphaFold 3
+      paper. Set this to false if providing custom paired MSA using the unpaired
+      MSA field to keep it exactly as is as deduplication against the paired MSA
+      could break the manually crafted pairing between MSA sequences.
+    verbose: Whether to print progress messages.
+
+  Returns:
+    A featurised batch for each rng_seed in the input.
+  """
+  validate_fold_input(fold_input)
+
+  # Set up data pipeline for single use.
+  data_pipeline = pipeline.WholePdbPipeline(
+      config=pipeline.WholePdbPipeline.Config(
+          buckets=buckets,
+          ref_max_modified_date=ref_max_modified_date,
+          conformer_max_iterations=conformer_max_iterations,
+          resolve_msa_overlaps=resolve_msa_overlaps,
+      ),
+  )
+
+  batches = []
+  for rng_seed in fold_input.rng_seeds:
+    featurisation_start_time = time.time()
+    if verbose:
+      print(f'Featurising data with seed {rng_seed}.')
+    batch = data_pipeline.process_item(
+        fold_input=fold_input,
+        ccd=ccd,
+        random_state=np.random.RandomState(rng_seed),
+        random_seed=rng_seed,
+    )
+    if verbose:
+      print(
+          f'Featurising data with seed {rng_seed} took'
+          f' {time.time() - featurisation_start_time:.2f} seconds.'
+      )
+    batches.append(batch)
+
+  return batches

flax_model/alphafold3/data/msa_config.py

@@ -0,0 +1,203 @@
+
+
+"""Genetic search config settings for data pipelines."""
+
+import dataclasses
+import datetime
+from typing import Self
+from flax_model.alphafold3.constants import mmcif_names
+
+
+def _validate_chain_poly_type(chain_poly_type: str) -> None:
+  if chain_poly_type not in mmcif_names.STANDARD_POLYMER_CHAIN_TYPES:
+    raise ValueError(
+        'chain_poly_type must be one of'
+        f' {mmcif_names.STANDARD_POLYMER_CHAIN_TYPES}: {chain_poly_type}'
+    )
+
+
+@dataclasses.dataclass(frozen=True, kw_only=True, slots=True)
+class DatabaseConfig:
+  """Configuration for a database."""
+
+  name: str
+  path: str
+
+
+@dataclasses.dataclass(frozen=True, kw_only=True, slots=True)
+class JackhmmerConfig:
+  """Configuration for a jackhmmer run.
+
+  Attributes:
+      binary_path: Path to the binary of the msa tool.
+      database_config: Database configuration.
+      n_cpu: An integer with the number of CPUs to use.
+      n_iter: An integer with the number of database search iterations.
+      e_value: e-value for the database lookup.
+      z_value: The Z-value representing the database size in number of sequences
+        for E-value and domain E-value calculation. Must be set for sharded
+        databases.
+      dom_z_value: The Z-value representing the database size in number of
+        sequences for domain E-value calculation. Must be set for sharded
+        databases.
+      max_sequences: Max sequences to return in MSA.
+      max_parallel_shards: If given, the maximum number of shards to search
+        against in parallel. If None, one Jackhmmer instance will be run per
+        shard. Only applicable if the database is sharded.
+      max_threads: If given, the maximum number of threads used when running
+        sharded databases.
+  """
+
+  binary_path: str
+  database_config: DatabaseConfig
+  n_cpu: int
+  n_iter: int
+  e_value: float
+  z_value: int | None
+  dom_z_value: int | None
+  max_sequences: int
+  max_parallel_shards: int | None = None
+  max_threads: int | None = None
+
+  
+@dataclasses.dataclass(frozen=True, kw_only=True, slots=True)
+class NhmmerConfig:
+  """Configuration for a nhmmer run.
+
+  Attributes:
+      binary_path: Path to the binary of the msa tool.
+      hmmalign_binary_path: Path to the hmmalign binary.
+      hmmbuild_binary_path: Path to the hmmbuild binary.
+      database_config: Database configuration.
+      n_cpu: An integer with the number of CPUs to use.
+      e_value: e-value for the database lookup.
+      z_value: The Z-value representing the database size in megabases for
+        E-value calculation. Allows fractional values. Must be set for sharded
+        databases.
+      max_sequences: Max sequences to return in MSA.
+      alphabet: The alphabet when building a profile with hmmbuild.
+      max_parallel_shards: If given, the maximum number of shards to search
+        against in parallel. If None, one Nhmmer instance will be run per shard.
+        Only applicable if the database is sharded.
+      max_threads: If given, the maximum number of threads used when running
+        sharded databases.        
+  """
+
+  binary_path: str
+  hmmalign_binary_path: str
+  hmmbuild_binary_path: str
+  database_config: DatabaseConfig
+  n_cpu: int
+  e_value: float
+  z_value: float | None
+  max_sequences: int
+  alphabet: str | None
+  max_parallel_shards: int | None = None
+  max_threads: int | None = None
+  
+
+@dataclasses.dataclass(frozen=True, kw_only=True, slots=True)
+class MmseqsConfig:
+  """Configuration for mmseqs run.
+
+  Attributes:
+      binary_path: Path to the binary of the msa tool.
+      database_config: Database configuration.
+      n_cpu: An integer with the number of CPUs to use.
+      use_gpu: An integer with the number of GPUs to use.
+      mmseqs_options: Options for the mmseqs run.
+      result2msa_options: Options for the result2msa run.
+  """
+
+  binary_path: str
+  database_config: DatabaseConfig
+  n_cpu: int
+  use_gpu: int=0
+  mmseqs_options: str
+  result2msa_options: str
+  max_sequences: int
+
+@dataclasses.dataclass(frozen=True, kw_only=True, slots=True)
+class RunConfig:
+  """Configuration for an MSA run.
+
+  Attributes:
+    config: MSA tool config.
+    chain_poly_type: The chain type for which the tools will be run.
+    crop_size: The maximum number of sequences to keep in the MSA. If None, all
+      sequences are kept. Note that the query is included in the MSA, so it
+      doesn't make sense to set this to less than 2.
+  """
+
+  config: JackhmmerConfig | NhmmerConfig | MmseqsConfig
+  chain_poly_type: str
+  crop_size: int | None
+
+  def __post_init__(self):
+    if self.crop_size is not None and self.crop_size < 2:
+      raise ValueError(f'crop_size must be None or >= 2: {self.crop_size}')
+
+    _validate_chain_poly_type(self.chain_poly_type)
+
+
+@dataclasses.dataclass(frozen=True, kw_only=True, slots=True)
+class HmmsearchConfig:
+  """Configuration for a hmmsearch."""
+
+  hmmsearch_binary_path: str
+  hmmbuild_binary_path: str
+
+  e_value: float
+  inc_e: float
+  dom_e: float
+  incdom_e: float
+  alphabet: str = 'amino'
+  filter_f1: float | None = None
+  filter_f2: float | None = None
+  filter_f3: float | None = None
+  filter_max: bool = False
+
+
+@dataclasses.dataclass(frozen=True, kw_only=True, slots=True)
+class TemplateToolConfig:
+  """Configuration for a template tool."""
+
+  database_path: str
+  chain_poly_type: str
+  hmmsearch_config: HmmsearchConfig
+  max_a3m_query_sequences: int | None = 300
+
+  def __post_init__(self):
+    _validate_chain_poly_type(self.chain_poly_type)
+
+
+@dataclasses.dataclass(frozen=True, kw_only=True, slots=True)
+class TemplateFilterConfig:
+  """Configuration for a template filter."""
+
+  max_subsequence_ratio: float | None
+  min_align_ratio: float | None
+  min_hit_length: int | None
+  deduplicate_sequences: bool
+  max_hits: int | None
+  max_template_date: datetime.date
+
+  @classmethod
+  def no_op_filter(cls) -> Self:
+    """Returns a config for filter that keeps everything."""
+    return cls(
+        max_subsequence_ratio=None,
+        min_align_ratio=None,
+        min_hit_length=None,
+        deduplicate_sequences=False,
+        max_hits=None,
+        max_template_date=datetime.date(3000, 1, 1),  # Very far in the future.
+    )
+
+
+@dataclasses.dataclass(frozen=True, kw_only=True, slots=True)
+class TemplatesConfig:
+  """Configuration for the template search pipeline."""
+
+  template_tool_config: TemplateToolConfig
+  filter_config: TemplateFilterConfig

flax_model/alphafold3/data/msa_features.py

@@ -0,0 +1,194 @@
+
+
+"""Utilities for computing MSA features."""
+
+from collections.abc import Sequence
+import re
+from flax_model.alphafold3.constants import mmcif_names
+import numpy as np
+
+_PROTEIN_TO_ID = {
+    'A': 0,
+    'B': 3,  # Same as D.
+    'C': 4,
+    'D': 3,
+    'E': 6,
+    'F': 13,
+    'G': 7,
+    'H': 8,
+    'I': 9,
+    'J': 20,  # Same as unknown (X).
+    'K': 11,
+    'L': 10,
+    'M': 12,
+    'N': 2,
+    'O': 20,  # Same as unknown (X).
+    'P': 14,
+    'Q': 5,
+    'R': 1,
+    'S': 15,
+    'T': 16,
+    'U': 4,  # Same as C.
+    'V': 19,
+    'W': 17,
+    'X': 20,
+    'Y': 18,
+    'Z': 6,  # Same as E.
+    '-': 21,
+}
+
+_RNA_TO_ID = {
+    # Map non-standard residues to UNK_NUCLEIC (N) -> 30
+    **{chr(i): 30 for i in range(ord('A'), ord('Z') + 1)},
+    # Continue the RNA indices from where Protein indices left off.
+    '-': 21,
+    'A': 22,
+    'G': 23,
+    'C': 24,
+    'U': 25,
+}
+
+_DNA_TO_ID = {
+    # Map non-standard residues to UNK_NUCLEIC (N) -> 30
+    **{chr(i): 30 for i in range(ord('A'), ord('Z') + 1)},
+    # Continue the DNA indices from where DNA indices left off.
+    '-': 21,
+    'A': 26,
+    'G': 27,
+    'C': 28,
+    'T': 29,
+}
+
+
+def extract_msa_features(
+    msa_sequences: Sequence[str], chain_poly_type: str
+) -> tuple[np.ndarray, np.ndarray]:
+  """Extracts MSA features.
+
+  Example:
+  The input raw MSA is: `[["AAAAAA"], ["Ai-CiDiiiEFa"]]`
+  The output MSA will be: `[["AAAAAA"], ["A-CDEF"]]`
+  The deletions will be: `[[0, 0, 0, 0, 0, 0], [0, 1, 0, 1, 3, 0]]`
+
+  Args:
+    msa_sequences: A list of strings, each string with one MSA sequence. Each
+      string must have the same, constant number of non-lowercase (matching)
+      residues.
+    chain_poly_type: Either 'polypeptide(L)' (protein), 'polyribonucleotide'
+      (RNA), or 'polydeoxyribonucleotide' (DNA). Use the appropriate string
+      constant from mmcif_names.py.
+
+  Returns:
+    A tuple with:
+    * MSA array of shape (num_seq, num_res) that contains only the uppercase
+      characters or gaps (-) from the original MSA.
+    * Deletions array of shape (num_seq, num_res) that contains the number
+      of deletions (lowercase letters in the MSA) to the left from each
+      non-deleted residue (uppercase letters in the MSA).
+
+  Raises:
+    ValueError if any of the preconditions are not met.
+  """
+
+  # Select the appropriate character map based on the chain type.
+  if chain_poly_type == mmcif_names.RNA_CHAIN:
+    char_map = _RNA_TO_ID
+  elif chain_poly_type == mmcif_names.DNA_CHAIN:
+    char_map = _DNA_TO_ID
+  elif chain_poly_type == mmcif_names.PROTEIN_CHAIN:
+    char_map = _PROTEIN_TO_ID
+  else:
+    raise ValueError(f'{chain_poly_type=} invalid.')
+
+  # Handle empty MSA.
+  if not msa_sequences:
+    empty_msa = np.array([], dtype=np.int32).reshape((0, 0))
+    empty_deletions = np.array([], dtype=np.int32).reshape((0, 0))
+    return empty_msa, empty_deletions
+
+  # Get the number of rows and columns in the MSA.
+  num_rows = len(msa_sequences)
+  num_cols = sum(1 for c in msa_sequences[0] if c in char_map)
+
+  # Initialize the output arrays.
+  msa_arr = np.zeros((num_rows, num_cols), dtype=np.int32)
+  deletions_arr = np.zeros((num_rows, num_cols), dtype=np.int32)
+
+  # Populate the output arrays.
+  for problem_row, msa_sequence in enumerate(msa_sequences):
+    deletion_count = 0
+    upper_count = 0
+    problem_col = 0
+    problems = []
+    for current in msa_sequence:
+      msa_id = char_map.get(current, -1)
+      if msa_id == -1:
+        if not current.islower():
+          problems.append(f'({problem_row}, {problem_col}):{current}')
+        deletion_count += 1
+      else:
+        # Check the access is safe before writing to the array.
+        # We don't need to check problem_row since it's guaranteed to be within
+        # the array bounds, while upper_count is incremented in the loop.
+        if upper_count < deletions_arr.shape[1]:
+          deletions_arr[problem_row, upper_count] = deletion_count
+          msa_arr[problem_row, upper_count] = msa_id
+        deletion_count = 0
+        upper_count += 1
+      problem_col += 1
+    if problems:
+      raise ValueError(
+          f"Unknown residues in MSA: {', '.join(problems)}. "
+          f'target_sequence: {msa_sequences[0]}'
+      )
+    if upper_count != num_cols:
+      raise ValueError(
+          'Invalid shape all strings must have the same number '
+          'of non-lowercase characters; First string has '
+          f"{num_cols} non-lowercase characters but '{msa_sequence}' has "
+          f'{upper_count}. target_sequence: {msa_sequences[0]}'
+      )
+
+  return msa_arr, deletions_arr
+
+
+# UniProtKB SwissProt/TrEMBL dbs have the following description format:
+# `db|UniqueIdentifier|EntryName`, e.g. `sp|P0C2L1|A3X1_LOXLA` or
+# `tr|A0A146SKV9|A0A146SKV9_FUNHE`.
+_UNIPROT_ENTRY_NAME_REGEX = re.compile(
+    # UniProtKB TrEMBL or SwissProt database.
+    r'(?:tr|sp)\|'
+    # A primary accession number of the UniProtKB entry.
+    r'(?:[A-Z0-9]{6,10})'
+    # Occasionally there is an isoform suffix (e.g. _1 or _10) which we ignore.
+    r'(?:_\d+)?\|'
+    # TrEMBL: Same as AccessionId (6-10 characters).
+    # SwissProt: A mnemonic protein identification code (1-5 characters).
+    r'(?:[A-Z0-9]{1,10}_)'
+    # A mnemonic species identification code.
+    r'(?P<SpeciesId>[A-Z0-9]{1,5})'
+)
+
+
+def extract_species_ids(msa_descriptions: Sequence[str]) -> Sequence[str]:
+  """Extracts species ID from MSA UniProtKB sequence identifiers.
+
+  Args:
+    msa_descriptions: The descriptions (the FASTA/A3M comment line) for each of
+      the sequences.
+
+  Returns:
+    Extracted UniProtKB species IDs if there is a regex match for each
+    description line, blank if the regex doesn't match.
+  """
+  species_ids = []
+  for msa_description in msa_descriptions:
+    msa_description = msa_description.strip()
+    match = _UNIPROT_ENTRY_NAME_REGEX.match(msa_description)
+    if match:
+      species_ids.append(match.group('SpeciesId'))
+    else:
+      # Handle cases where the regex doesn't match
+      # (e.g., append None or raise an error depending on your needs)
+      species_ids.append('')
+  return species_ids

flax_model/alphafold3/data/msa_identifiers.py

@@ -0,0 +1,77 @@
+
+
+"""Utilities for extracting identifiers from MSA sequence descriptions."""
+
+import dataclasses
+import re
+
+
+# Sequences coming from UniProtKB database come in the
+# `db|UniqueIdentifier|EntryName` format, e.g. `tr|A0A146SKV9|A0A146SKV9_FUNHE`
+# or `sp|P0C2L1|A3X1_LOXLA` (for TREMBL/Swiss-Prot respectively).
+_UNIPROT_PATTERN = re.compile(
+    r"""
+    ^
+    # UniProtKB/TrEMBL or UniProtKB/Swiss-Prot
+    (?:tr|sp)
+    \|
+    # A primary accession number of the UniProtKB entry.
+    (?P<AccessionIdentifier>[A-Za-z0-9]{6,10})
+    # Occasionally there is a _0 or _1 isoform suffix, which we ignore.
+    (?:_\d)?
+    \|
+    # TREMBL repeats the accession ID here. Swiss-Prot has a mnemonic
+    # protein ID code.
+    (?:[A-Za-z0-9]+)
+    _
+    # A mnemonic species identification code.
+    (?P<SpeciesIdentifier>([A-Za-z0-9]){1,5})
+    # Small BFD uses a final value after an underscore, which we ignore.
+    (?:_\d+)?
+    $
+    """,
+    re.VERBOSE,
+)
+
+
+@dataclasses.dataclass(frozen=True)
+class Identifiers:
+  species_id: str = ''
+
+
+def _parse_sequence_identifier(msa_sequence_identifier: str) -> Identifiers:
+  """Gets species from an msa sequence identifier.
+
+  The sequence identifier has the format specified by
+  _UNIPROT_TREMBL_ENTRY_NAME_PATTERN or _UNIPROT_SWISSPROT_ENTRY_NAME_PATTERN.
+  An example of a sequence identifier: `tr|A0A146SKV9|A0A146SKV9_FUNHE`
+
+  Args:
+    msa_sequence_identifier: a sequence identifier.
+
+  Returns:
+    An `Identifiers` instance with species_id. These
+    can be empty in the case where no identifier was found.
+  """
+  matches = re.search(_UNIPROT_PATTERN, msa_sequence_identifier.strip())
+  if matches:
+    return Identifiers(species_id=matches.group('SpeciesIdentifier'))
+  return Identifiers()
+
+
+def _extract_sequence_identifier(description: str) -> str | None:
+  """Extracts sequence identifier from description. Returns None if no match."""
+  split_description = description.split()
+  if split_description:
+    return split_description[0].partition('/')[0]
+  else:
+    return None
+
+
+def get_identifiers(description: str) -> Identifiers:
+  """Computes extra MSA features from the description."""
+  sequence_identifier = _extract_sequence_identifier(description)
+  if sequence_identifier is None:
+    return Identifiers()
+  else:
+    return _parse_sequence_identifier(sequence_identifier)

flax_model/alphafold3/data/parsers.py

@@ -0,0 +1,261 @@
+
+
+"""Functions for parsing various file formats."""
+
+from collections.abc import Iterable, Sequence
+from typing import IO, TypeAlias
+
+from flax_model.alphafold3.cpp import fasta_iterator
+from flax_model.alphafold3.cpp import msa_conversion
+
+
+DeletionMatrix: TypeAlias = Sequence[Sequence[int]]
+
+
+def lazy_parse_fasta_string(fasta_string: str) -> Iterable[tuple[str, str]]:
+  """Lazily parses a FASTA/A3M string and yields (sequence, description) tuples.
+
+  This implementation is more memory friendly than `fasta_sequence` while
+  offering comparable performance. The underlying implementation is in C++ and
+  is therefore faster than a pure Python implementation.
+
+  Use this method when parsing FASTA files where you already have the FASTA
+  string, but need to control how far you iterate through its sequences.
+
+  Arguments:
+    fasta_string: A string with the contents of FASTA/A3M file.
+
+  Returns:
+    Iterator of (sequence, description). In the description, the leading ">" is
+    stripped.
+
+  Raises:
+    ValueError if the FASTA/A3M file is invalid, e.g. empty.
+  """
+
+  # The lifetime of the FastaStringIterator is tied to the lifetime of
+  # fasta_string - fasta_string must be kept while the iterator is in use.
+  return fasta_iterator.FastaStringIterator(fasta_string)
+
+
+def parse_fasta(fasta_string: str) -> tuple[Sequence[str], Sequence[str]]:
+  """Parses FASTA string and returns list of strings with amino-acid sequences.
+
+  Arguments:
+    fasta_string: The string contents of a FASTA file.
+
+  Returns:
+    A tuple of two lists:
+    * A list of sequences.
+    * A list of sequence descriptions taken from the comment lines. In the
+      same order as the sequences.
+  """
+  return fasta_iterator.parse_fasta_include_descriptions(fasta_string)
+
+
+def convert_a3m_to_stockholm(a3m: str, max_seqs: int | None = None) -> str:
+  """Converts MSA in the A3M format to the Stockholm format."""
+  sequences, descriptions = parse_fasta(a3m)
+  if max_seqs is not None:
+    sequences = sequences[:max_seqs]
+    descriptions = descriptions[:max_seqs]
+
+  stockholm = ['# STOCKHOLM 1.0', '']
+
+  # Add the Stockholm header with the sequence metadata.
+  names = []
+  for i, description in enumerate(descriptions):
+    name, _, rest = description.replace('\t', ' ').partition(' ')
+    # Ensure that the names are unique - stockholm format requires that
+    # the sequence names are unique.
+    name = f'{name}_{i}'
+    names.append(name)
+    # Avoid zero-length description due to historic hmmbuild parsing bug.
+    desc = rest.strip() or '<EMPTY>'
+    stockholm.append(f'#=GS {name.strip()} DE {desc}')
+  stockholm.append('')
+
+  # Convert insertions in a sequence into gaps in all other sequences that don't
+  # have an insertion in that column as well.
+  sequences = msa_conversion.convert_a3m_to_stockholm(sequences)
+
+  # Add the MSA data.
+  max_name_width = max(len(name) for name in names)
+  for name, sequence in zip(names, sequences, strict=True):
+    # Align the names to the left and pad with spaces to the maximum length.
+    stockholm.append(f'{name:<{max_name_width}s} {sequence}')
+
+  # Add the reference annotation for the query (the first sequence).
+  ref_annotation = ''.join('.' if c == '-' else 'x' for c in sequences[0])
+  stockholm.append(f'{"#=GC RF":<{max_name_width}s} {ref_annotation}')
+  stockholm.append('//')
+
+  return '\n'.join(stockholm)
+
+
+def convert_stockholm_to_a3m(
+    stockholm: IO[str],
+    max_sequences: int | None = None,
+    remove_first_row_gaps: bool = True,
+    linewidth: int | None = None,
+) -> str:
+  """Converts MSA in Stockholm format to the A3M format."""
+  descriptions = {}
+  sequences = {}
+  reached_max_sequences = False
+
+  if linewidth is not None and linewidth <= 0:
+    raise ValueError('linewidth must be > 0 or None')
+
+  for line in stockholm:
+    reached_max_sequences = max_sequences and len(sequences) >= max_sequences
+    line = line.strip()
+    # Ignore blank lines, markup and end symbols - remainder are alignment
+    # sequence parts.
+    if not line or line.startswith(('#', '//')):
+      continue
+    seqname, aligned_seq = line.split(maxsplit=1)
+    if seqname not in sequences:
+      if reached_max_sequences:
+        continue
+      sequences[seqname] = ''
+    sequences[seqname] += aligned_seq
+
+  if not sequences:
+    return ''
+
+  stockholm.seek(0)
+  for line in stockholm:
+    line = line.strip()
+    if line[:4] == '#=GS':
+      # Description row - example format is:
+      # #=GS UniRef90_Q9H5Z4/4-78            DE [subseq from] cDNA: FLJ22755 ...
+      columns = line.split(maxsplit=3)
+      seqname, feature = columns[1:3]
+      value = columns[3] if len(columns) == 4 else ''
+      if feature != 'DE':
+        continue
+      if reached_max_sequences and seqname not in sequences:
+        continue
+      descriptions[seqname] = value
+      if len(descriptions) == len(sequences):
+        break
+
+  assert len(descriptions) <= len(sequences)
+
+  # Convert sto format to a3m line by line
+  a3m_sequences = {}
+  # query_sequence is assumed to be the first sequence
+  query_sequence = next(iter(sequences.values()))
+  for seqname, sto_sequence in sequences.items():
+    if remove_first_row_gaps:
+      a3m_sequences[seqname] = msa_conversion.align_sequence_to_gapless_query(
+          sequence=sto_sequence, query_sequence=query_sequence
+      ).replace('.', '')
+    else:
+      a3m_sequences[seqname] = sto_sequence.replace('.', '')
+
+  fasta_chunks = []
+
+  for seqname, seq in a3m_sequences.items():
+    fasta_chunks.append(f'>{seqname} {descriptions.get(seqname, "")}')
+
+    if linewidth:
+      fasta_chunks.extend(
+          seq[i : linewidth + i] for i in range(0, len(seq), linewidth)
+      )
+    else:
+      fasta_chunks.append(seq)
+
+  return '\n'.join(fasta_chunks) + '\n'  # Include terminating newline.
+
+
+def convert_mmseqs_stockholm_to_a3m(
+    stockholm: IO[str],
+    max_sequences: int | None = None,
+    remove_first_row_gaps: bool = True,
+    linewidth: int | None = None,
+) -> str:
+    """Converts MSA in Stockholm format to the A3M format."""
+    from collections import defaultdict
+
+    descriptions = {}
+    sequences = {}  # 存储最终序列（自动处理重复键）
+    seqname_counter = defaultdict(int)  # 记录原始seqname出现次数
+    original_seqnames = {}  # 记录处理后的seqname对应的原始名�?
+
+    if linewidth is not None and linewidth <= 0:
+        raise ValueError('linewidth must be > 0 or None')
+
+    # 第一遍：读取所有序列行，处理重复键
+    for line in stockholm:
+        line = line.strip()
+        if not line or line.startswith(('#', '//')):
+            continue
+        parts = line.split(maxsplit=1)
+        if len(parts) < 2:
+            continue
+        original_seqname, aligned_seq = parts
+
+        # 生成唯一seqname（例�?"_1", "_1_a", "_1_b"�?
+        count = seqname_counter[original_seqname]
+        if count == 0:
+            new_seqname = original_seqname
+        else:
+            new_seqname = f"{original_seqname}_{chr(97 + count - 1)}"  # 97是ASCII码的'a'
+        seqname_counter[original_seqname] += 1
+
+        # 达到最大序列数时跳过后续序�?
+        if max_sequences and len(sequences) >= max_sequences:
+            continue
+
+        # 累积序列（原逻辑�?
+        sequences.setdefault(new_seqname, '')
+        sequences[new_seqname] += aligned_seq
+        original_seqnames[new_seqname] = original_seqname
+
+    if not sequences:
+        return ''
+
+    # 第二遍：读取描述信息
+    stockholm.seek(0)
+    for line in stockholm:
+        line = line.strip()
+        if line.startswith('#=GS'):
+            columns = line.split(maxsplit=3)
+            if len(columns) < 4:
+                continue
+            seqname, feature = columns[1:3]
+            if feature != 'DE':
+                continue
+            value = columns[3] if len(columns) >= 4 else ''
+            descriptions[seqname] = value
+
+    # 转换序列格式
+    a3m_sequences = {}
+    query_sequence = next(iter(sequences.values())).replace('.', '')  # 假设第一个序列是查询序列
+
+    for seqname, sto_sequence in sequences.items():
+        current_seq = sto_sequence.replace('.', '')
+        if remove_first_row_gaps:
+            aligned = msa_conversion.align_sequence_to_gapless_query(
+                sequence=current_seq, query_sequence=query_sequence
+            ).replace('.', '')
+        else:
+            aligned = current_seq
+        a3m_sequences[seqname] = aligned
+
+    # 生成FASTA
+    fasta_chunks = []
+    for seqname, seq in a3m_sequences.items():
+        original_seqname = original_seqnames.get(seqname, seqname)
+        desc = descriptions.get(original_seqname, "")
+        fasta_chunks.append(f'>{seqname} {desc}')
+
+        if linewidth:
+            chunks = [seq[i:i+linewidth] for i in range(0, len(seq), linewidth)]
+            fasta_chunks.extend(chunks)
+        else:
+            fasta_chunks.append(seq)
+
+    return '\n'.join(fasta_chunks) + '\n'

flax_model/alphafold3/data/pipeline.py

@@ -0,0 +1,677 @@
+
+
+"""Functions for running the MSA and template tools for the AlphaFold model."""
+
+from concurrent import futures
+import dataclasses
+import datetime
+import functools
+import logging
+import time
+
+from flax_model.alphafold3.common import folding_input
+from flax_model.alphafold3.constants import mmcif_names
+from flax_model.alphafold3.data import msa
+from flax_model.alphafold3.data import msa_config
+from flax_model.alphafold3.data import structure_stores
+from flax_model.alphafold3.data import templates as templates_lib
+
+
+
+# Cache to avoid re-running template search for the same sequence in homomers.
+@functools.cache
+def _get_protein_templates(
+    sequence: str,
+    input_msa_a3m: str,
+    run_template_search: bool,
+    templates_config: msa_config.TemplatesConfig,
+    pdb_database_path: str,
+) -> templates_lib.Templates:
+  """Searches for templates for a single protein chain."""
+  if run_template_search:
+    templates_start_time = time.time()
+    logging.info('Getting protein templates for sequence %s', sequence)
+    protein_templates = templates_lib.Templates.from_seq_and_a3m(
+        query_sequence=sequence,
+        msa_a3m=input_msa_a3m,
+        max_template_date=templates_config.filter_config.max_template_date,
+        database_path=templates_config.template_tool_config.database_path,
+        hmmsearch_config=templates_config.template_tool_config.hmmsearch_config,
+        max_a3m_query_sequences=None,
+        chain_poly_type=mmcif_names.PROTEIN_CHAIN,
+        structure_store=structure_stores.StructureStore(pdb_database_path),
+        filter_config=templates_config.filter_config,
+    )
+    logging.info(
+        'Getting %d protein templates took %.2f seconds for sequence %s',
+        protein_templates.num_hits,
+        time.time() - templates_start_time,
+        sequence,
+    )
+  else:
+    logging.info('Skipping template search for sequence %s', sequence)
+    protein_templates = templates_lib.Templates(
+        query_sequence=sequence,
+        hits=[],
+        max_template_date=templates_config.filter_config.max_template_date,
+        structure_store=structure_stores.StructureStore(pdb_database_path),
+    )
+  return protein_templates
+
+
+# Cache to avoid re-running the MSA tools for the same sequence in homomers.
+@functools.cache
+def _get_protein_msa_and_templates(
+    sequence: str,
+    run_template_search: bool,
+    uniref90_msa_config: msa_config.RunConfig,
+    mgnify_msa_config: msa_config.RunConfig,
+    small_bfd_msa_config: msa_config.RunConfig,
+    uniprot_msa_config: msa_config.RunConfig,
+    templates_config: msa_config.TemplatesConfig,
+    pdb_database_path: str,
+    model_loader_callback=None,
+) -> tuple[msa.Msa, msa.Msa, templates_lib.Templates]:
+  """Processes a single protein chain."""
+  logging.info('Getting protein MSAs for sequence %s', sequence)
+  msa_start_time = time.time()
+  # Run various MSA tools in parallel. Use a ThreadPoolExecutor because
+  # they're not blocked by the GIL, as they're sub-shelled out.
+  with futures.ThreadPoolExecutor(max_workers=4) as executor:
+    uniref90_msa_future = executor.submit(
+        msa.get_msa,
+        target_sequence=sequence,
+        run_config=uniref90_msa_config,
+        chain_poly_type=mmcif_names.PROTEIN_CHAIN,
+    )
+    mgnify_msa_future = executor.submit(
+        msa.get_msa,
+        target_sequence=sequence,
+        run_config=mgnify_msa_config,
+        chain_poly_type=mmcif_names.PROTEIN_CHAIN,
+    )
+    small_bfd_msa_future = executor.submit(
+        msa.get_msa,
+        target_sequence=sequence,
+        run_config=small_bfd_msa_config,
+        chain_poly_type=mmcif_names.PROTEIN_CHAIN,
+    )
+    uniprot_msa_future = executor.submit(
+        msa.get_msa,
+        target_sequence=sequence,
+        run_config=uniprot_msa_config,
+        chain_poly_type=mmcif_names.PROTEIN_CHAIN,
+    )
+  uniref90_msa = uniref90_msa_future.result()
+  mgnify_msa = mgnify_msa_future.result()
+  small_bfd_msa = small_bfd_msa_future.result()
+  uniprot_msa = uniprot_msa_future.result()
+  logging.info(
+      'Getting protein MSAs took %.2f seconds for sequence %s',
+      time.time() - msa_start_time,
+      sequence,
+  )
+  
+  # Load model after protein MSAs are completed
+  if model_loader_callback is not None:
+    logging.info('Protein MSAs completed. Loading model parameters...')
+    model_loader_callback()
+
+  logging.info('Deduplicating MSAs for sequence %s', sequence)
+  msa_dedupe_start_time = time.time()
+  with futures.ThreadPoolExecutor() as executor:
+    unpaired_protein_msa_future = executor.submit(
+        msa.Msa.from_multiple_msas,
+        msas=[uniref90_msa, small_bfd_msa, mgnify_msa],
+        deduplicate=True,
+    )
+    paired_protein_msa_future = executor.submit(
+        msa.Msa.from_multiple_msas, msas=[uniprot_msa], deduplicate=False
+    )
+  unpaired_protein_msa = unpaired_protein_msa_future.result()
+  paired_protein_msa = paired_protein_msa_future.result()
+  logging.info(
+      'Deduplicating MSAs took %.2f seconds for sequence %s, found %d unpaired'
+      ' sequences, %d paired sequences',
+      time.time() - msa_dedupe_start_time,
+      sequence,
+      unpaired_protein_msa.depth,
+      paired_protein_msa.depth,
+  )
+
+  protein_templates = _get_protein_templates(
+      sequence=sequence,
+      input_msa_a3m=unpaired_protein_msa.to_a3m(),
+      run_template_search=run_template_search,
+      templates_config=templates_config,
+      pdb_database_path=pdb_database_path,
+  )
+
+  return unpaired_protein_msa, paired_protein_msa, protein_templates
+
+
+# Cache to avoid re-running the Nhmmer for the same sequence in homomers.
+@functools.cache
+def _get_rna_msa(
+    sequence: str,
+    nt_rna_msa_config: msa_config.NhmmerConfig,
+    rfam_msa_config: msa_config.NhmmerConfig,
+    rnacentral_msa_config: msa_config.NhmmerConfig,
+) -> msa.Msa:
+  """Processes a single RNA chain."""
+  logging.info('Getting RNA MSAs for sequence %s', sequence)
+  rna_msa_start_time = time.time()
+  # Run various MSA tools in parallel. Use a ThreadPoolExecutor because
+  # they're not blocked by the GIL, as they're sub-shelled out.
+  with futures.ThreadPoolExecutor() as executor:
+    nt_rna_msa_future = executor.submit(
+        msa.get_msa,
+        target_sequence=sequence,
+        run_config=nt_rna_msa_config,
+        chain_poly_type=mmcif_names.RNA_CHAIN,
+    )
+    rfam_msa_future = executor.submit(
+        msa.get_msa,
+        target_sequence=sequence,
+        run_config=rfam_msa_config,
+        chain_poly_type=mmcif_names.RNA_CHAIN,
+    )
+    rnacentral_msa_future = executor.submit(
+        msa.get_msa,
+        target_sequence=sequence,
+        run_config=rnacentral_msa_config,
+        chain_poly_type=mmcif_names.RNA_CHAIN,
+    )
+  nt_rna_msa = nt_rna_msa_future.result()
+  rfam_msa = rfam_msa_future.result()
+  rnacentral_msa = rnacentral_msa_future.result()
+  rna_msa = msa.Msa.from_multiple_msas(
+      msas=[rfam_msa, rnacentral_msa, nt_rna_msa],
+      deduplicate=True,
+  )
+  logging.info(
+      'Getting RNA MSAs took %.2f seconds for sequence %s, found %d unpaired'
+      ' sequences',
+      time.time() - rna_msa_start_time,
+      sequence,
+      rna_msa.depth,
+  )
+  return rna_msa
+
+
+@dataclasses.dataclass(frozen=True, slots=True, kw_only=True)
+class DataPipelineConfig:
+  """The configuration for the data pipeline.
+
+  Attributes:
+    jackhmmer_binary_path: Jackhmmer binary path, used for protein MSA search.
+    nhmmer_binary_path: Nhmmer binary path, used for RNA MSA search.
+    hmmalign_binary_path: Hmmalign binary path, used to align hits to the query
+      profile.
+    hmmsearch_binary_path: Hmmsearch binary path, used for template search.
+    hmmbuild_binary_path: Hmmbuild binary path, used to build HMM profile from
+      raw MSA in template search.
+    small_bfd_database_path: Small BFD database path, used for protein MSA
+      search.
+    small_bfd_z_value: The Z-value representing the database size in number of
+      sequences for E-value calculation. Must be set for sharded databases.
+    mgnify_database_path: Mgnify database path, used for protein MSA search.
+    mgnify_z_value: The Z-value representing the database size in number of
+      sequences for E-value calculation. Must be set for sharded databases.
+    uniprot_cluster_annot_database_path: Uniprot database path, used for protein
+      paired MSA search.
+    uniprot_cluster_annot_z_value: The Z-value representing the database size in
+      number of sequences for E-value calculation. Must be set for sharded
+      databases.
+    uniref90_database_path: UniRef90 database path, used for MSA search, and the
+      MSA obtained by searching it is used to construct the profile for template
+      search.
+    uniref90_z_value: The Z-value representing the database size in number of
+      sequences for E-value calculation. Must be set for sharded databases.
+    ntrna_database_path: NT-RNA database path, used for RNA MSA search.
+    ntrna_z_value: The Z-value representing the database size in megabases for
+      E-value calculation. Must be set for sharded databases.
+    rfam_database_path: Rfam database path, used for RNA MSA search.
+    rfam_z_value: The Z-value representing the database size in megabases for
+      E-value calculation. Must be set for sharded databases.
+    rna_central_database_path: RNAcentral database path, used for RNA MSA
+      search.
+    rna_central_z_value: The Z-value representing the database size in megabases
+      for E-value calculation. Must be set for sharded databases.
+    seqres_database_path: PDB sequence database path, used for template search.
+    pdb_database_path: PDB database directory with mmCIF files path, used for
+      template search.
+    jackhmmer_n_cpu: Number of CPUs to use for Jackhmmer.
+    jackhmmer_max_parallel_shards: Maximum number of shards to search against in
+      parallel. If None, one Jackhmmer instance will be run per shard. Only
+      applicable if the database is sharded.
+    nhmmer_n_cpu: Number of CPUs to use for Nhmmer.
+    nhmmer_max_parallel_shards: Maximum number of shards to search against in
+      parallel. If None, one Nhmmer instance will be run per shard. Only
+      applicable if the database is sharded.
+    max_template_date: The latest date of templates to use.
+  """
+
+  # Binary paths.
+  jackhmmer_binary_path: str
+  nhmmer_binary_path: str
+  hmmalign_binary_path: str
+  hmmsearch_binary_path: str
+  hmmbuild_binary_path: str
+  mmseqs_binary_path: str
+
+  # Jackhmmer databases.
+  small_bfd_database_path: str
+  small_bfd_z_value: int | None = None
+  mgnify_database_path: str
+  mgnify_z_value: int | None = None
+  uniprot_cluster_annot_database_path: str
+  uniprot_cluster_annot_z_value: int | None = None
+  uniref90_database_path: str
+  uniref90_z_value: int | None = None
+  # Nhmmer databases.
+  ntrna_database_path: str
+  ntrna_z_value: int | None = None
+  rfam_database_path: str
+  rfam_z_value: int | None = None
+  rna_central_database_path: str
+  rna_central_z_value: int | None = None
+  # Template search databases.
+  seqres_database_path: str
+  pdb_database_path: str
+
+  # Optional configuration for MSA tools.
+  jackhmmer_n_cpu: int = 8
+  jackhmmer_max_parallel_shards: int | None = None
+  jackhmmer_max_threads: int | None = None
+  nhmmer_n_cpu: int = 8
+  nhmmer_max_parallel_shards: int | None = None
+  nhmmer_max_threads: int | None = None
+  mmseqs_n_cpu: int = 8
+
+  max_template_date: datetime.date
+
+  use_mmseqs: bool = False
+  use_mmseqs_gpu: bool = False
+  mmseqs_options: str
+  result2msa_options: str
+
+class DataPipeline:
+  """Runs the alignment tools and assembles the input features."""
+
+  def __init__(self, data_pipeline_config: DataPipelineConfig, model_loader_callback=None):
+    """Initializes the data pipeline with default configurations."""
+    if not data_pipeline_config.use_mmseqs:
+      self._uniref90_msa_config = msa_config.RunConfig(
+          config=msa_config.JackhmmerConfig(
+              binary_path=data_pipeline_config.jackhmmer_binary_path,
+              database_config=msa_config.DatabaseConfig(
+                  name='uniref90',
+                  path=data_pipeline_config.uniref90_database_path,
+              ),
+              n_cpu=data_pipeline_config.jackhmmer_n_cpu,
+              n_iter=1,
+              e_value=1e-4,
+              z_value=data_pipeline_config.uniref90_z_value,
+              dom_z_value=data_pipeline_config.uniref90_z_value,
+              max_sequences=10_000,
+              max_parallel_shards=data_pipeline_config.jackhmmer_max_parallel_shards,
+              max_threads=data_pipeline_config.jackhmmer_max_threads,
+          ),
+          chain_poly_type=mmcif_names.PROTEIN_CHAIN,
+          crop_size=None,
+      )
+      self._mgnify_msa_config = msa_config.RunConfig(
+          config=msa_config.JackhmmerConfig(
+              binary_path=data_pipeline_config.jackhmmer_binary_path,
+              database_config=msa_config.DatabaseConfig(
+                  name='mgnify',
+                  path=data_pipeline_config.mgnify_database_path,
+              ),
+              n_cpu=data_pipeline_config.jackhmmer_n_cpu,
+              n_iter=1,
+              e_value=1e-4,
+              z_value=data_pipeline_config.mgnify_z_value,
+              dom_z_value=data_pipeline_config.mgnify_z_value,
+              max_sequences=5_000,
+              max_parallel_shards=data_pipeline_config.jackhmmer_max_parallel_shards,
+              max_threads=data_pipeline_config.jackhmmer_max_threads,
+          ),
+          chain_poly_type=mmcif_names.PROTEIN_CHAIN,
+          crop_size=None,
+      )
+      self._small_bfd_msa_config = msa_config.RunConfig(
+          config=msa_config.JackhmmerConfig(
+              binary_path=data_pipeline_config.jackhmmer_binary_path,
+              database_config=msa_config.DatabaseConfig(
+                  name='small_bfd',
+                  path=data_pipeline_config.small_bfd_database_path,
+              ),
+              n_cpu=data_pipeline_config.jackhmmer_n_cpu,
+              n_iter=1,
+              e_value=1e-4,
+              # Set z_value=138_515_945 to match the z_value used in the paper.
+              # In practice, this has minimal impact on predicted structures.
+              z_value=data_pipeline_config.small_bfd_z_value,
+              dom_z_value=data_pipeline_config.small_bfd_z_value,
+              max_sequences=5_000,
+              max_parallel_shards=data_pipeline_config.jackhmmer_max_parallel_shards,
+              max_threads=data_pipeline_config.jackhmmer_max_threads,
+          ),
+          chain_poly_type=mmcif_names.PROTEIN_CHAIN,
+          crop_size=None,
+      )
+      self._uniprot_msa_config = msa_config.RunConfig(
+          config=msa_config.JackhmmerConfig(
+              binary_path=data_pipeline_config.jackhmmer_binary_path,
+              database_config=msa_config.DatabaseConfig(
+                  name='uniprot_cluster_annot',
+                  path=data_pipeline_config.uniprot_cluster_annot_database_path,
+              ),
+              n_cpu=data_pipeline_config.jackhmmer_n_cpu,
+              n_iter=1,
+              e_value=1e-4,
+              z_value=data_pipeline_config.uniprot_cluster_annot_z_value,
+              dom_z_value=data_pipeline_config.uniprot_cluster_annot_z_value,
+              max_sequences=50_000,
+              max_parallel_shards=data_pipeline_config.jackhmmer_max_parallel_shards,
+              max_threads=data_pipeline_config.jackhmmer_max_threads,
+          ),
+          chain_poly_type=mmcif_names.PROTEIN_CHAIN,
+          crop_size=None,
+      )
+    else:
+      self._uniref90_msa_config = msa_config.RunConfig(
+          config=msa_config.MmseqsConfig(
+              binary_path=data_pipeline_config.mmseqs_binary_path,
+              database_config=msa_config.DatabaseConfig(
+                    name='uniref90',
+                    path=data_pipeline_config.uniref90_database_path,
+              ),
+              n_cpu=data_pipeline_config.mmseqs_n_cpu,
+              use_gpu=1 if data_pipeline_config.use_mmseqs_gpu else 0,
+              mmseqs_options=data_pipeline_config.mmseqs_options,
+              result2msa_options=data_pipeline_config.result2msa_options,
+              max_sequences=10_000,
+          ),
+          chain_poly_type=mmcif_names.PROTEIN_CHAIN,
+          crop_size=None,
+      )
+      self._mgnify_msa_config = msa_config.RunConfig(
+          config=msa_config.MmseqsConfig(
+              binary_path=data_pipeline_config.mmseqs_binary_path,
+              database_config=msa_config.DatabaseConfig(
+                  name='mgnify',
+                  path=data_pipeline_config.mgnify_database_path,
+              ),
+              n_cpu=data_pipeline_config.mmseqs_n_cpu,
+              use_gpu=1 if data_pipeline_config.use_mmseqs_gpu else 0,
+              mmseqs_options=data_pipeline_config.mmseqs_options,
+              result2msa_options=data_pipeline_config.result2msa_options,
+              max_sequences=5_000,
+          ),
+          chain_poly_type=mmcif_names.PROTEIN_CHAIN,
+          crop_size=None,
+      )
+      self._small_bfd_msa_config = msa_config.RunConfig(
+          config=msa_config.MmseqsConfig(
+              binary_path=data_pipeline_config.mmseqs_binary_path,
+              database_config=msa_config.DatabaseConfig(
+                  name='small_bfd',
+                  path=data_pipeline_config.small_bfd_database_path,
+              ),
+              n_cpu=data_pipeline_config.mmseqs_n_cpu,
+              use_gpu=1 if data_pipeline_config.use_mmseqs_gpu else 0,
+              mmseqs_options=data_pipeline_config.mmseqs_options,
+              result2msa_options=data_pipeline_config.result2msa_options,
+              max_sequences=5_000,
+          ),
+          chain_poly_type=mmcif_names.PROTEIN_CHAIN,
+          crop_size=None,
+      )
+      self._uniprot_msa_config = msa_config.RunConfig(
+          config=msa_config.MmseqsConfig(
+              binary_path=data_pipeline_config.mmseqs_binary_path,
+              database_config=msa_config.DatabaseConfig(
+                  name='uniprot_cluster_annot',
+                  path=data_pipeline_config.uniprot_cluster_annot_database_path,
+              ),
+              n_cpu=data_pipeline_config.mmseqs_n_cpu,
+              use_gpu=1 if data_pipeline_config.use_mmseqs_gpu else 0,
+              mmseqs_options=data_pipeline_config.mmseqs_options,
+              result2msa_options=data_pipeline_config.result2msa_options,
+              max_sequences=50_000,
+          ),
+          chain_poly_type=mmcif_names.PROTEIN_CHAIN,
+          crop_size=None,
+      )
+    
+    self._nt_rna_msa_config = msa_config.RunConfig(
+        config=msa_config.NhmmerConfig(
+            binary_path=data_pipeline_config.nhmmer_binary_path,
+            hmmalign_binary_path=data_pipeline_config.hmmalign_binary_path,
+            hmmbuild_binary_path=data_pipeline_config.hmmbuild_binary_path,
+            database_config=msa_config.DatabaseConfig(
+                name='nt_rna',
+                path=data_pipeline_config.ntrna_database_path,
+            ),
+            n_cpu=data_pipeline_config.nhmmer_n_cpu,
+            e_value=1e-3,
+            alphabet='rna',
+            z_value=data_pipeline_config.ntrna_z_value,
+            max_sequences=10_000,
+            max_parallel_shards=data_pipeline_config.nhmmer_max_parallel_shards,
+            max_threads=data_pipeline_config.nhmmer_max_threads,
+        ),
+        chain_poly_type=mmcif_names.RNA_CHAIN,
+        crop_size=None,
+    )
+    self._rfam_msa_config = msa_config.RunConfig(
+        config=msa_config.NhmmerConfig(
+            binary_path=data_pipeline_config.nhmmer_binary_path,
+            hmmalign_binary_path=data_pipeline_config.hmmalign_binary_path,
+            hmmbuild_binary_path=data_pipeline_config.hmmbuild_binary_path,
+            database_config=msa_config.DatabaseConfig(
+                name='rfam_rna',
+                path=data_pipeline_config.rfam_database_path,
+            ),
+            n_cpu=data_pipeline_config.nhmmer_n_cpu,
+            e_value=1e-3,
+            alphabet='rna',
+            z_value=data_pipeline_config.rfam_z_value,
+            max_sequences=10_000,
+            max_parallel_shards=data_pipeline_config.nhmmer_max_parallel_shards,
+            max_threads=data_pipeline_config.nhmmer_max_threads,
+        ),
+        chain_poly_type=mmcif_names.RNA_CHAIN,
+        crop_size=None,
+    )
+    self._rnacentral_msa_config = msa_config.RunConfig(
+        config=msa_config.NhmmerConfig(
+            binary_path=data_pipeline_config.nhmmer_binary_path,
+            hmmalign_binary_path=data_pipeline_config.hmmalign_binary_path,
+            hmmbuild_binary_path=data_pipeline_config.hmmbuild_binary_path,
+            database_config=msa_config.DatabaseConfig(
+                name='rna_central_rna',
+                path=data_pipeline_config.rna_central_database_path,
+            ),
+            n_cpu=data_pipeline_config.nhmmer_n_cpu,
+            e_value=1e-3,
+            alphabet='rna',
+            z_value=data_pipeline_config.rna_central_z_value,
+            max_sequences=10_000,
+            max_parallel_shards=data_pipeline_config.nhmmer_max_parallel_shards,
+            max_threads=data_pipeline_config.nhmmer_max_threads,
+        ),
+        chain_poly_type=mmcif_names.RNA_CHAIN,
+        crop_size=None,
+    )
+
+    self._templates_config = msa_config.TemplatesConfig(
+        template_tool_config=msa_config.TemplateToolConfig(
+            database_path=data_pipeline_config.seqres_database_path,
+            chain_poly_type=mmcif_names.PROTEIN_CHAIN,
+            hmmsearch_config=msa_config.HmmsearchConfig(
+                hmmsearch_binary_path=data_pipeline_config.hmmsearch_binary_path,
+                hmmbuild_binary_path=data_pipeline_config.hmmbuild_binary_path,
+                filter_f1=0.1,
+                filter_f2=0.1,
+                filter_f3=0.1,
+                e_value=100,
+                inc_e=100,
+                dom_e=100,
+                incdom_e=100,
+                alphabet='amino',
+            ),
+        ),
+        filter_config=msa_config.TemplateFilterConfig(
+            max_subsequence_ratio=0.95,
+            min_align_ratio=0.1,
+            min_hit_length=10,
+            deduplicate_sequences=True,
+            max_hits=4,
+            max_template_date=data_pipeline_config.max_template_date,
+        ),
+    )
+    self._pdb_database_path = data_pipeline_config.pdb_database_path
+    self._model_loader_callback = model_loader_callback
+
+  def process_protein_chain(
+      self, chain: folding_input.ProteinChain
+  ) -> folding_input.ProteinChain:
+    """Processes a single protein chain."""
+    has_unpaired_msa = chain.unpaired_msa is not None
+    has_paired_msa = chain.paired_msa is not None
+    has_templates = chain.templates is not None
+
+    if not has_unpaired_msa and not has_paired_msa and not chain.templates:
+      # MSA None - search. Templates either [] - don't search, or None - search.
+      unpaired_msa, paired_msa, template_hits = _get_protein_msa_and_templates(
+          sequence=chain.sequence,
+          run_template_search=not has_templates,  # Skip template search if [].
+          uniref90_msa_config=self._uniref90_msa_config,
+          mgnify_msa_config=self._mgnify_msa_config,
+          small_bfd_msa_config=self._small_bfd_msa_config,
+          uniprot_msa_config=self._uniprot_msa_config,
+          templates_config=self._templates_config,
+          pdb_database_path=self._pdb_database_path,
+          model_loader_callback=self._model_loader_callback,
+      )
+      unpaired_msa = unpaired_msa.to_a3m()
+      paired_msa = paired_msa.to_a3m()
+      templates = [
+          folding_input.Template(
+              mmcif=struc.to_mmcif(),
+              query_to_template_map=hit.query_to_hit_mapping,
+          )
+          for hit, struc in template_hits.get_hits_with_structures()
+      ]
+    elif has_unpaired_msa and has_paired_msa and not has_templates:
+      # Has MSA, but doesn't have templates. Search for templates only.
+      empty_msa = msa.Msa.from_empty(
+          query_sequence=chain.sequence,
+          chain_poly_type=mmcif_names.PROTEIN_CHAIN,
+      ).to_a3m()
+      unpaired_msa = chain.unpaired_msa or empty_msa
+      paired_msa = chain.paired_msa or empty_msa
+      template_hits = _get_protein_templates(
+          sequence=chain.sequence,
+          input_msa_a3m=unpaired_msa,
+          run_template_search=True,
+          templates_config=self._templates_config,
+          pdb_database_path=self._pdb_database_path,
+      )
+      templates = [
+          folding_input.Template(
+              mmcif=struc.to_mmcif(),
+              query_to_template_map=hit.query_to_hit_mapping,
+          )
+          for hit, struc in template_hits.get_hits_with_structures()
+      ]
+    else:
+      # Has MSA and templates, don't search for anything.
+      if not has_unpaired_msa or not has_paired_msa or not has_templates:
+        raise ValueError(
+            f'Protein chain {chain.id} has unpaired MSA, paired MSA, or'
+            ' templates set only partially. If you want to run the pipeline'
+            ' with custom MSA/templates, you need to set all of them. You can'
+            ' set MSA to empty string and templates to empty list to signify'
+            ' that they should not be used and searched for.'
+        )
+      logging.info(
+          'Skipping MSA and template search for protein chain %s because it '
+          'already has MSAs and templates.',
+          chain.id,
+      )
+      if not chain.unpaired_msa:
+        logging.info('Using empty unpaired MSA for protein chain %s', chain.id)
+      if not chain.paired_msa:
+        logging.info('Using empty paired MSA for protein chain %s', chain.id)
+      if not chain.templates:
+        logging.info('Using no templates for protein chain %s', chain.id)
+      empty_msa = msa.Msa.from_empty(
+          query_sequence=chain.sequence,
+          chain_poly_type=mmcif_names.PROTEIN_CHAIN,
+      ).to_a3m()
+      unpaired_msa = chain.unpaired_msa or empty_msa
+      paired_msa = chain.paired_msa or empty_msa
+      templates = chain.templates
+
+    return folding_input.ProteinChain(
+        id=chain.id,
+        sequence=chain.sequence,
+        ptms=chain.ptms,
+        unpaired_msa=unpaired_msa,
+        paired_msa=paired_msa,
+        templates=templates,
+    )
+
+  def process_rna_chain(
+      self, chain: folding_input.RnaChain
+  ) -> folding_input.RnaChain:
+    """Processes a single RNA chain."""
+    if chain.unpaired_msa is not None:
+      # Don't run MSA tools if the chain already has an MSA.
+      logging.info(
+          'Skipping MSA search for RNA chain %s because it already has MSA.',
+          chain.id,
+      )
+      if not chain.unpaired_msa:
+        logging.info('Using empty unpaired MSA for RNA chain %s', chain.id)
+      empty_msa = msa.Msa.from_empty(
+          query_sequence=chain.sequence, chain_poly_type=mmcif_names.RNA_CHAIN
+      ).to_a3m()
+      unpaired_msa = chain.unpaired_msa or empty_msa
+    else:
+      unpaired_msa = _get_rna_msa(
+          sequence=chain.sequence,
+          nt_rna_msa_config=self._nt_rna_msa_config,
+          rfam_msa_config=self._rfam_msa_config,
+          rnacentral_msa_config=self._rnacentral_msa_config,
+      ).to_a3m()
+    return folding_input.RnaChain(
+        id=chain.id,
+        sequence=chain.sequence,
+        modifications=chain.modifications,
+        unpaired_msa=unpaired_msa,
+    )
+
+  def process(self, fold_input: folding_input.Input) -> folding_input.Input:
+    """Runs MSA and template tools and returns a new Input with the results."""
+    processed_chains = []
+    for chain in fold_input.chains:
+      print(f'Running data pipeline for chain {chain.id}...')
+      process_chain_start_time = time.time()
+      match chain:
+        case folding_input.ProteinChain():
+          processed_chains.append(self.process_protein_chain(chain))
+        case folding_input.RnaChain():
+          processed_chains.append(self.process_rna_chain(chain))
+        case _:
+          processed_chains.append(chain)
+      print(
+          f'Running data pipeline for chain {chain.id} took'
+          f' {time.time() - process_chain_start_time:.2f} seconds',
+      )
+
+    return dataclasses.replace(fold_input, chains=processed_chains)

flax_model/alphafold3/data/structure_stores.py

@@ -0,0 +1,92 @@
+
+
+"""Library for loading structure data from various sources."""
+
+from collections.abc import Mapping, Sequence
+import functools
+import os
+import pathlib
+import tarfile
+
+
+class NotFoundError(KeyError):
+  """Raised when the structure store doesn't contain the requested target."""
+
+
+class StructureStore:
+  """Handles the retrieval of mmCIF files from a filesystem."""
+
+  def __init__(
+      self,
+      structures: str | os.PathLike[str] | Mapping[str, str],
+  ):
+    """Initialises the instance.
+
+    Args:
+      structures: Path of the directory where the mmCIF files are or a Mapping
+        from target name to mmCIF string.
+    """
+    if isinstance(structures, Mapping):
+      self._structure_mapping = structures
+      self._structure_path = None
+      self._structure_tar = None
+    else:
+      self._structure_mapping = None
+      path_str = os.fspath(structures)
+      if path_str.endswith('.tar'):
+        self._structure_tar = tarfile.open(path_str, 'r')
+        self._structure_path = None
+      else:
+        self._structure_path = pathlib.Path(structures)
+        self._structure_tar = None
+
+  @functools.cached_property
+  def _tar_members(self) -> Mapping[str, tarfile.TarInfo]:
+    assert self._structure_tar is not None
+    return {
+        path.stem: tarinfo
+        for tarinfo in self._structure_tar.getmembers()
+        if tarinfo.isfile()
+        and (path := pathlib.Path(tarinfo.path.lower())).suffix == '.cif'
+    }
+
+  def get_mmcif_str(self, target_name: str) -> str:
+    """Returns an mmCIF for a given `target_name`.
+
+    Args:
+      target_name: Name specifying the target mmCIF.
+
+    Raises:
+      NotFoundError: If the target is not found.
+    """
+    if self._structure_mapping is not None:
+      try:
+        return self._structure_mapping[target_name]
+      except KeyError as e:
+        raise NotFoundError(f'{target_name=} not found') from e
+
+    if self._structure_tar is not None:
+      try:
+        member = self._tar_members[target_name]
+        if struct_file := self._structure_tar.extractfile(member):
+          return struct_file.read().decode()
+        else:
+          raise NotFoundError(f'{target_name=} not found')
+      except KeyError:
+        raise NotFoundError(f'{target_name=} not found') from None
+
+    filepath = self._structure_path / f'{target_name}.cif'
+    try:
+      return filepath.read_text()
+    except FileNotFoundError as e:
+      raise NotFoundError(f'{target_name=} not found at {filepath=}') from e
+
+  def target_names(self) -> Sequence[str]:
+    """Returns all targets in the store."""
+    if self._structure_mapping is not None:
+      return [*self._structure_mapping.keys()]
+    elif self._structure_tar is not None:
+      return sorted(self._tar_members.keys())
+    elif self._structure_path is not None:
+      return sorted([path.stem for path in self._structure_path.glob('*.cif')])
+    return ()

flax_model/alphafold3/data/template_store.py

@@ -0,0 +1,38 @@
+
+
+"""Interface and implementations for fetching templates data."""
+
+from collections.abc import Mapping
+import datetime
+from typing import Any, Protocol, TypeAlias
+
+
+TemplateFeatures: TypeAlias = Mapping[str, Any]
+
+
+class TemplateFeatureProvider(Protocol):
+  """Interface for providing Template Features."""
+
+  def __call__(
+      self,
+      sequence: str,
+      release_date: datetime.date | None,
+      include_ligand_features: bool = True,
+  ) -> TemplateFeatures:
+    """Retrieve template features for the given sequence and release_date.
+
+    Args:
+      sequence: The residue sequence of the query.
+      release_date: The release_date of the template query, this is used to
+        filter templates for training, ensuring that they do not leak structure
+        information from the future.
+      include_ligand_features: Whether to include ligand features.
+
+    Returns:
+      Template features: A mapping of template feature labels to features, which
+        may be numpy arrays, bytes objects, or for the special case of label
+        `ligand_features`, a nested feature map of labels to numpy arrays.
+
+    Raises:
+      TemplateRetrievalError if the template features were not found.
+    """

flax_model/alphafold3/data/tools/msa_tool.py

@@ -0,0 +1,31 @@
+
+
+"""Defines protocol for MSA tools."""
+
+import dataclasses
+from typing import Protocol
+
+
+@dataclasses.dataclass(frozen=True, slots=True, kw_only=True)
+class MsaToolResult:
+  """The result of a MSA tool query.
+
+  Attributes:
+    target_sequence: The sequence that was used to query the MSA tool.
+    e_value: The e-value that was used to filter the MSA tool results.
+    a3m: The MSA output of the tool in the A3M format.
+    tblout: The optional tblout output of the MSA tool (needed for merging
+      results of queries against a sharded database).
+  """
+
+  target_sequence: str
+  e_value: float
+  a3m: str
+  tblout: str | None = None
+
+
+class MsaTool(Protocol):
+  """Interface for MSA tools."""
+
+  def query(self, target_sequence: str) -> MsaToolResult:
+    """Runs the MSA tool on the target sequence."""

flax_model/alphafold3/data/tools/nhmmer.py

@@ -0,0 +1,351 @@
+
+
+"""Library to run Nhmmer from Python."""
+
+from collections.abc import Iterable, Sequence
+from concurrent import futures
+import heapq
+import os
+import pathlib
+import shutil
+import tempfile
+import time
+from typing import Final
+
+from absl import logging
+from flax_model.alphafold3.data import parsers
+from flax_model.alphafold3.data.tools import hmmalign
+from flax_model.alphafold3.data.tools import hmmbuild
+from flax_model.alphafold3.data.tools import msa_tool
+from flax_model.alphafold3.data.tools import shards
+from flax_model.alphafold3.data.tools import subprocess_utils
+
+_SHORT_SEQUENCE_CUTOFF: Final[int] = 50
+
+
+class Nhmmer(msa_tool.MsaTool):
+  """Python wrapper of the Nhmmer binary."""
+
+  def __init__(
+      self,
+      binary_path: str,
+      hmmalign_binary_path: str,
+      hmmbuild_binary_path: str,
+      database_path: str,
+      n_cpu: int = 8,
+      e_value: float = 1e-3,
+      z_value: float | int | None = None,
+      max_sequences: int = 5000,
+      filter_f3: float = 1e-5,
+      alphabet: str | None = None,
+      strand: str | None = None,
+      max_threads: int | None = None,
+  ):
+    """Initializes the Python Nhmmer wrapper.
+
+    NOTE: The MSA obtained by running against sharded dbs won't be always
+    exactly the same as the MSA obtained by running against an unsharded db.
+    This is because of Jackhmmer deduplication logic, which won't spot duplicate
+    hits across multiple shards. Usually this means that the sharded search
+    finds more hits (likely bounded by the number of shards), but this should
+    not pose an issue given how the results are used downstream. The problem is
+    more pronounced with deep MSAs and lower in the hit list (higher e-values).
+
+    Make sure to set the Z value when searching against a sharded database,
+    otherwise the results won't match the normal unsharded search.
+
+    Args:
+      binary_path: Path to the Nhmmer binary.
+      hmmalign_binary_path: Path to the Hmmalign binary.
+      hmmbuild_binary_path: Path to the Hmmbuild binary.
+      database_path: MSA database path to search against. This can be either a
+        FASTA (slow) or HMMERDB produced from the FASTA using the makehmmerdb
+        binary. The HMMERDB is ~10x faster but experimental.  Sharded file
+        specs, e.g. <db_path>@<num_shards>, are supported.
+      n_cpu: The number of CPUs to give Nhmmer.
+      e_value: The E-value, see Nhmmer docs for more details. Will be
+        overwritten if bit_score is set.
+      z_value: The Z-value representing the number of comparisons done (i.e
+        correct database size) for E-value calculation. Make sure to set this
+        when searching against a sharded database, otherwise the e-values will
+        be incorrectly scaled.
+      max_sequences: Maximum number of sequences to return in the MSA.
+      filter_f3: Forward pre-filter, set to >1.0 to turn off.
+      alphabet: The alphabet to assert when building a profile with hmmbuild.
+        This must be 'rna', 'dna', or None.
+      strand: "watson" searches query sequence, "crick" searches
+        reverse-compliment and default is None which means searching for both.
+      max_threads: If given, the maximum number of threads used when running
+        sharded databases.
+
+    Raises:
+      RuntimeError: If Nhmmer binary not found within the path.
+      ValueError: If an invalid configuration is provided in the args.
+    """
+    self._database_path = database_path
+
+    if shard_paths := shards.get_sharded_paths(self._database_path):
+      if z_value is None:
+        raise ValueError(
+            'The Z-value must be set when searching against a sharded database '
+            'to correctly scale e-values.'
+        )
+      if 'hmmerdb' in self._database_path:
+        raise ValueError('HMMERDB is not supported in sharded mode.')
+
+      if max_sequences <= 1:
+        raise ValueError(
+            'max_sequences must be greater than 1 when running in sharded '
+            'mode, because each shard would return only the query sequence.'
+        )
+
+      self._shard_paths = shard_paths
+      self._max_threads = len(self._shard_paths)
+      if max_threads is not None:
+        self._max_threads = min(max_threads, self._max_threads)
+      logging.info('Nhmmer running with max_threads = %d', self._max_threads)
+    else:
+      self._shard_paths = None
+      self._max_threads = None
+
+    self._binary_path = binary_path
+    self._hmmalign_binary_path = hmmalign_binary_path
+    self._hmmbuild_binary_path = hmmbuild_binary_path
+    subprocess_utils.check_binary_exists(path=self._binary_path, name='Nhmmer')
+
+    if strand and strand not in {'watson', 'crick'}:
+      raise ValueError(f'Invalid {strand=}. only "watson" or "crick" supported')
+
+    if alphabet and alphabet not in {'rna', 'dna'}:
+      raise ValueError(f'Invalid {alphabet=}, only "rna" or "dna" supported')
+
+    self._e_value = e_value
+    self._n_cpu = n_cpu
+    self._z_value = z_value
+    self._max_sequences = max_sequences
+    self._filter_f3 = filter_f3
+    self._alphabet = alphabet
+    self._strand = strand
+
+  def query(self, target_sequence: str) -> msa_tool.MsaToolResult:
+    """Query the database (sharded or unsharded) using Nhmmer."""
+    if self._shard_paths:
+      # Sharded case, run the query against each database shard in parallel.
+      logging.info(
+          'Query sequence (sharded db): %s',
+          target_sequence
+          if len(target_sequence) <= 16
+          else f'{target_sequence[:16]}... (len {len(target_sequence)})',
+      )
+
+      global_temp_dir = tempfile.mkdtemp()
+
+      def _query_shard_fn(
+          shard_path: str,
+      ) -> tuple[msa_tool.MsaToolResult, float]:
+        t_start = time.time()
+        # Get tblout as it contains e-values we need for merging sequences.
+        result = self._query_db_shard(
+            target_sequence=target_sequence,
+            db_shard_path=shard_path,
+            get_tblout=True,  # Tblout contains e-values needed for merging.
+            global_temp_dir=global_temp_dir,
+        )
+        return result, time.time() - t_start
+
+      with futures.ThreadPoolExecutor(max_workers=self._max_threads) as ex:
+        tool_outputs, timings = zip(*ex.map(_query_shard_fn, self._shard_paths))
+
+      logging.info(
+          'Finished query for %d shards, shard timings (seconds): %s',
+          len(tool_outputs),
+          ', '.join(f'{t:.1f}' for t in timings),
+      )
+
+      shutil.rmtree(global_temp_dir, ignore_errors=True)
+      return _merge_nhmmer_results(tool_outputs, self._max_sequences)
+
+    else:
+      # Non-sharded case, run the query against the whole database.
+      logging.info(
+          'Query sequence (non-sharded db): %s',
+          target_sequence
+          if len(target_sequence) <= 16
+          else f'{target_sequence[:16]}... (len {len(target_sequence)})',
+      )
+      return self._query_db_shard(
+          target_sequence=target_sequence,
+          db_shard_path=self._database_path,
+          get_tblout=False,
+      )
+
+  def _query_db_shard(
+      self,
+      *,
+      target_sequence: str,
+      db_shard_path: str,
+      get_tblout: bool,
+      global_temp_dir: str | None = None,
+  ) -> msa_tool.MsaToolResult:
+    """Query the database shard using Nhmmer."""
+
+    with tempfile.TemporaryDirectory(dir=global_temp_dir) as query_tmp_dir:
+      input_a3m_path = os.path.join(query_tmp_dir, 'query.a3m')
+      output_sto_path = os.path.join(query_tmp_dir, 'output.sto')
+      pathlib.Path(output_sto_path).touch()
+      subprocess_utils.create_query_fasta_file(
+          sequence=target_sequence, path=input_a3m_path
+      )
+
+      cmd_flags = [
+          *('-o', '/dev/null'),  # Don't pollute stdout with nhmmer output.
+          '--noali',  # Don't include the alignment in stdout.
+          *('--cpu', str(self._n_cpu)),
+      ]
+
+      if get_tblout:
+        output_tblout_path = pathlib.Path(query_tmp_dir, 'tblout.txt')
+        output_tblout_path.touch()
+        cmd_flags.extend(['--tblout', str(output_tblout_path)])
+      else:
+        output_tblout_path = None
+
+      cmd_flags.extend(['-E', str(self._e_value)])
+
+      if self._z_value is not None:
+        cmd_flags.extend(['-Z', str(self._z_value)])
+
+      if self._alphabet:
+        cmd_flags.extend([f'--{self._alphabet}'])
+
+      if self._strand is not None:
+        cmd_flags.extend([f'--{self._strand}'])
+
+      cmd_flags.extend(['-A', output_sto_path])
+      # As recommend by RNAcentral for short sequences.
+      if (
+          self._alphabet == 'rna'
+          and len(target_sequence) < _SHORT_SEQUENCE_CUTOFF
+      ):
+        cmd_flags.extend(['--F3', str(0.02)])
+      else:
+        cmd_flags.extend(['--F3', str(self._filter_f3)])
+
+      # The input A3M and the db are the last two arguments.
+      cmd_flags.extend((input_a3m_path, db_shard_path))
+
+      cmd = [self._binary_path, *cmd_flags]
+      subprocess_utils.run(
+          cmd=cmd,
+          cmd_name=f'Nhmmer ({os.path.basename(db_shard_path)})',
+          log_stdout=False,
+          log_stderr=True,
+          log_on_process_error=True,
+      )
+
+      if os.path.getsize(output_sto_path) > 0:
+        with open(output_sto_path) as f:
+          a3m_out = parsers.convert_stockholm_to_a3m(
+              f, max_sequences=self._max_sequences - 1  # Query not included.
+          )
+        # Nhmmer hits are generally shorter than the query sequence. To get MSA
+        # of width equal to the query sequence, align hits to the query profile.
+        logging.info('Aligning output a3m of size %d bytes', len(a3m_out))
+
+        aligner = hmmalign.Hmmalign(self._hmmalign_binary_path)
+        target_sequence_fasta = f'>query\n{target_sequence}\n'
+        profile_builder = hmmbuild.Hmmbuild(
+            binary_path=self._hmmbuild_binary_path, alphabet=self._alphabet
+        )
+        profile = profile_builder.build_profile_from_a3m(target_sequence_fasta)
+        a3m_out = aligner.align_sequences_to_profile(
+            profile=profile, sequences_a3m=a3m_out
+        )
+        a3m_out = ''.join([target_sequence_fasta, a3m_out])
+
+        # Parse the output a3m to remove line breaks.
+        a3m = '\n'.join(
+            [f'>{n}\n{s}' for s, n in parsers.lazy_parse_fasta_string(a3m_out)]
+        )
+      else:
+        # Nhmmer returns an empty file if there are no hits.
+        # In this case return only the query sequence.
+        a3m = f'>query\n{target_sequence}'
+
+      # Get the tabular output which has e.g. e-value for each target.
+      tbl = '' if output_tblout_path is None else output_tblout_path.read_text()
+
+    return msa_tool.MsaToolResult(
+        target_sequence=target_sequence,
+        e_value=self._e_value,
+        a3m=a3m,
+        tblout=tbl,
+    )
+
+
+def _merge_nhmmer_results(
+    nhmmer_results: Sequence[msa_tool.MsaToolResult],
+    max_sequences: int,
+) -> msa_tool.MsaToolResult:
+  """Merges nhmmer result protos into a single one."""
+  assert len(set(nh_res.target_sequence for nh_res in nhmmer_results)) == 1
+  assert len(set(nh_res.e_value for nh_res in nhmmer_results)) == 1
+
+  # Parse the TBL output, create a mapping from unique hit ID to TBL line.
+  parsed_tbl = {}
+  for nhmmer_result in nhmmer_results:
+    assert nhmmer_result.tblout is not None
+    for line in nhmmer_result.tblout.splitlines():
+      if not line.startswith('#'):
+        line_fields = line.split(maxsplit=15)
+        accession = line_fields[0]
+        alignment_from = line_fields[6]
+        alignment_to = line_fields[7]
+        # This is the unique ID that is used in the output A3M.
+        unique_id = f'{accession}/{alignment_from}-{alignment_to}'
+        parsed_tbl[unique_id] = line
+
+  # Create an iterator and merge a3m info with tbl info.
+  def _merged_a3m_tbl_iter(a3m: str) -> Iterable[tuple[str, str, str, str]]:
+    # Don't parse the entire a3m, lazily parse only as many sequences as needed.
+    iterator = iter(parsers.lazy_parse_fasta_string(a3m))
+    next(iterator)  # Skip the query which isn't present in tblout.
+    for sequence, description in iterator:
+      name = description.partition(' ')[0]
+      if tbl_info := parsed_tbl.get(name):
+        # Skip sequences for which we don't have tbl information.
+        yield sequence, description, tbl_info, name
+
+  def sort_key(seq_data: tuple[str, str, str, str]) -> tuple[float, str]:
+    unused_seq, unused_description, tbl_info, name = seq_data
+    # Nucleic tblout has 16 space delimited columns. "-" used if no value
+    # present. We want e-value in column 12, so do only 13 splits. Use the name
+    # in case of an e-value tie.
+    return float(tbl_info.split(maxsplit=13)[12]), name
+
+  # A3M/TBL is sorted by e-value and name, hence we can merge them efficiently.
+  merged_a3m_and_tblout = heapq.merge(
+      *[_merged_a3m_tbl_iter(res.a3m) for res in nhmmer_results],
+      key=sort_key,
+  )
+
+  # Truncate the a3m to max_sequences. Do not truncate the tblout.
+  merged_tblout = []
+  merged_a3m = [f'>query\n{nhmmer_results[0].target_sequence}']
+  for seq, description, tbl_info, _ in merged_a3m_and_tblout:
+    merged_tblout.append(tbl_info)
+    if len(merged_a3m) < max_sequences:
+      merged_a3m.append(f'>{description}\n{seq}')
+
+  logging.info(
+      'Limiting merged MSA depth from %d to %d',
+      len(merged_tblout),
+      max_sequences,
+  )
+
+  return msa_tool.MsaToolResult(
+      target_sequence=nhmmer_results[0].target_sequence,
+      a3m='\n'.join(merged_a3m),
+      e_value=nhmmer_results[0].e_value,
+      tblout=None,  # We no longer need the tblout.
+  )

flax_model/alphafold3/jax/attention/attention_call_arg_specs.py

@@ -0,0 +1,54 @@
+
+
+"""Attention call argument specifications.
+
+Attention argument specifications used by users of the library.
+They are the most important test cases, and also cases for optimize
+performance of via autotuning.
+"""
+
+from typing import Any
+
+import jax
+
+ShapedArray = jax.ShapeDtypeStruct
+
+
+def _make_argspec(
+    *,
+    q_shape,
+    dtype,
+    k_shape=None,
+    v_shape=None,
+    bias_shape=None,
+    mask_shape=None,
+    **kwargs,
+) -> dict[str, Any]:
+  """Make argspec from shapes and kwargs."""
+  if k_shape is None:
+    k_shape = q_shape
+  if v_shape is None:
+    v_shape = k_shape
+
+  return dict(
+      query=ShapedArray(q_shape, dtype),
+      key=ShapedArray(k_shape, dtype),
+      value=ShapedArray(v_shape, dtype),
+      bias=ShapedArray(bias_shape, dtype) if bias_shape is not None else None,
+      mask=ShapedArray(mask_shape, 'bool_') if mask_shape is not None else None,
+      **kwargs,
+  )
+
+
+# A subset of the full set of argument specifications. Useful for tap-tests and
+# microbenchmarks.
+CALL_ARG_SPECS = dict(
+    vanilla_f32=_make_argspec(q_shape=(8, 1024, 4, 128), dtype='float32'),
+    vanilla_bf16=_make_argspec(q_shape=(8, 1024, 4, 128), dtype='bfloat16'),
+    alphafold=_make_argspec(
+        q_shape=(384, 384, 4, 32),
+        bias_shape=(1, 4, 384, 384),
+        mask_shape=(384, 1, 1, 384),
+        dtype='bfloat16',
+    ),
+)

flax_model/alphafold3/model/merging_features.py

@@ -0,0 +1,84 @@
+
+
+"""Methods for merging existing features to create a new example.
+
+Covers:
+- Merging features across chains.
+- Merging the paired and unpaired parts of the MSA.
+"""
+
+from typing import TypeAlias
+
+from flax_model.alphafold3.model import data_constants
+import jax.numpy as jnp
+import numpy as np
+
+NUM_SEQ_NUM_RES_MSA_FEATURES = data_constants.NUM_SEQ_NUM_RES_MSA_FEATURES
+NUM_SEQ_MSA_FEATURES = data_constants.NUM_SEQ_MSA_FEATURES
+MSA_PAD_VALUES = data_constants.MSA_PAD_VALUES
+
+
+xnp_ndarray: TypeAlias = np.ndarray | jnp.ndarray  # pylint: disable=invalid-name
+BatchDict: TypeAlias = dict[str, xnp_ndarray]
+
+
+def _pad_features_to_max(feat_name: str, chains: list[BatchDict], axis: int):
+  """Pad a set of features to the maximum size amongst all chains.
+
+  Args:
+    feat_name: The feature name to pad.
+    chains: A list of chains with associated features.
+    axis: Which axis to pad to the max.
+
+  Returns:
+    A list of features, all with the same size on the given axis.
+  """
+  max_num_seq = np.max([chain[feat_name].shape[axis] for chain in chains])
+
+  padded_feats = []
+  for chain in chains:
+    feat = chain[feat_name]
+
+    padding = np.zeros_like(feat.shape)  # pytype: disable=attribute-error
+    padding[axis] = max_num_seq - feat.shape[axis]  # pytype: disable=attribute-error
+    padding = [(0, p) for p in padding]
+    padded_feats.append(
+        np.pad(
+            feat,
+            padding,
+            mode='constant',
+            constant_values=MSA_PAD_VALUES[feat_name],
+        )
+    )
+  return padded_feats
+
+
+def merge_msa_features(feat_name: str, chains: list[BatchDict]) -> np.ndarray:
+  """Merges MSA features with shape (NUM_SEQ, NUM_RES) across chains."""
+  expected_dtype = chains[0][feat_name].dtype
+  if '_all_seq' in feat_name:
+    return np.concatenate(
+        [c.get(feat_name, np.array([], expected_dtype)) for c in chains], axis=1
+    )
+  else:
+    # Since each MSA can be of different lengths, we first need to pad them
+    # all to the size of the largest MSA before concatenating.
+    padded_feats = _pad_features_to_max(feat_name, chains, axis=0)
+    return np.concatenate(padded_feats, axis=1)
+
+
+def merge_paired_and_unpaired_msa(example: BatchDict) -> BatchDict:
+  """Concatenates the paired (all_seq) MSA features with the unpaired ones."""
+  new_example = dict(example)
+
+  for feature_name in NUM_SEQ_NUM_RES_MSA_FEATURES + NUM_SEQ_MSA_FEATURES:
+    if feature_name in example and feature_name + '_all_seq' in example:
+      feat = example[feature_name]
+      feat_all_seq = example[feature_name + '_all_seq']
+      merged_feat = np.concatenate([feat_all_seq, feat], axis=0)
+      new_example[feature_name] = merged_feat
+
+  new_example['num_alignments'] = np.array(
+      new_example['msa'].shape[0], dtype=np.int32
+  )
+  return new_example