lhallee commited on
Commit
7b02892
·
verified ·
1 Parent(s): 9f70ca8

protein_annotation_compendium_v1: the compendium of UniProtKB 2026_03

Browse files
.gitattributes CHANGED
@@ -58,3 +58,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
 
 
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
61
+ receipts/split.json filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,200 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ pretty_name: "protein annotation compendium v1"
3
+ license: other
4
+ task_categories:
5
+ - other
6
+ tags:
7
+ - protein
8
+ - biology
9
+ - multilabel
10
+ - synthyra
11
+ size_categories:
12
+ - 1M<n<10M
13
+ configs:
14
+ - config_name: whole_vocabulary
15
+ default: true
16
+ data_files: [{split: train, path: train.parquet}, {split: validation, path: validation.parquet}, {split: test, path: test.parquet}, {split: calibration, path: calibration.parquet}]
17
+ - config_name: go_bp
18
+ data_files: [{split: train, path: go_bp/train.parquet}, {split: validation, path: go_bp/validation.parquet}, {split: test, path: go_bp/test.parquet}, {split: calibration, path: go_bp/calibration.parquet}]
19
+ - config_name: go_mf
20
+ data_files: [{split: train, path: go_mf/train.parquet}, {split: validation, path: go_mf/validation.parquet}, {split: test, path: go_mf/test.parquet}, {split: calibration, path: go_mf/calibration.parquet}]
21
+ - config_name: go_cc
22
+ data_files: [{split: train, path: go_cc/train.parquet}, {split: validation, path: go_cc/validation.parquet}, {split: test, path: go_cc/test.parquet}, {split: calibration, path: go_cc/calibration.parquet}]
23
+ - config_name: localization
24
+ data_files: [{split: train, path: localization/train.parquet}, {split: validation, path: localization/validation.parquet}, {split: test, path: localization/test.parquet}, {split: calibration, path: localization/calibration.parquet}]
25
+ - config_name: ec
26
+ data_files: [{split: train, path: ec/train.parquet}, {split: validation, path: ec/validation.parquet}, {split: test, path: ec/test.parquet}, {split: calibration, path: ec/calibration.parquet}]
27
+ ---
28
+
29
+ # protein annotation compendium v1
30
+
31
+ Proteins of UniProtKB 2026_03, Swiss-Prot and the TrEMBL entries of Swiss-Prot quality, each with every annotation its sources state: GO, EC, Rhea, cofactors, locations, topology, keywords, features with positions, modifications, InterPro and Pfam families, Reactome pathways and lineage, each with its evidence code and one of four tiers. A row of the default config is one distinct sequence with its vocabulary tokens, the terms and tiers beside them; GO, localization and EC views and every pooled assertion ship beside it. Split 85/5/8/2 by cluster at 0.5 identity: 1,423,992 train, 18,680 validation, 31,662 test and 6,830 calibration proteins, no held-out protein at 0.5 identity or more to another split; the default config holds the 1,481,135 with a vocabulary term.
32
+
33
+ | | |
34
+ |---|---|
35
+ | Task | prediction of a protein's annotation terms from its sequence over one vocabulary in eight fundamental aspects (Oracle One, the translator), retrieval between a sequence and its annotation (CAMP), and per-aspect views for GO, localization and EC |
36
+ | Rows | 1,481,135 across four splits |
37
+ | Homology | enforced at 0.5 identity, measured at 0.3 |
38
+ | Acceptance | 5 of 5 acceptance checks passed, among them 12 directed homology searches with none at 0.5 identity or more |
39
+ | Status | private |
40
+
41
+ ## Homology leakage
42
+
43
+ **Enforced: 0.5 identity.** No held-out sequence has a relative at 0.5 identity or more in another split: every cluster at that level, and so every finer one, sits in exactly one split. 12 directed searches between the splits found no pair above 0.499 identity. **Measured, not enforced: 0.3 identity.** 18,767 of the 57,172 held-out proteins (33%) have no relative at 0.3 identity or more in train; the rest have one between 0.3 and 0.5.
44
+
45
+ Clustered at 0.9, 0.7, 0.5, 0.3 identity with 0.8 coverage of both sequences (MMseqs2).
46
+
47
+ | Identity | Guarantee |
48
+ |---|---|
49
+ | 0.9 | implied: no cluster at this level spans two splits |
50
+ | 0.7 | implied: no cluster at this level spans two splits |
51
+ | 0.5 | **enforced**: no cluster at this level spans two splits |
52
+ | 0.3 | measured, not enforced: 18,767 of 57,172 held-out proteins are below this identity to train |
53
+
54
+ | Split | Rows | Clusters (0.9 identity) |
55
+ |---|---|---|
56
+ | train | 1,423,964 | 798,403 |
57
+ | validation | 18,680 | 12,446 |
58
+ | test | 31,661 | 21,515 |
59
+ | calibration | 6,830 | 5,167 |
60
+
61
+ <details><summary>Why this level</summary>
62
+
63
+ CAMP, which reads this split, may enforce 0.5 at most, and no level up to it is well formed, so the survey of the clusters recommends 0.5, the finest such level: there 886 of the 6,066 labels spread over 20 or more units at 0.9 identity fall to fewer (14.6 percent, over the survey's 10), as their proteins sit in a few large families; 0.7 is the lowest well-formed level. Repair returned to train every held-out cluster with a relative in another split at 0.5 identity, since the clusters join 0.8 of both sequences and the audit counts half of one.
64
+
65
+ </details>
66
+
67
+ ## How it was built
68
+
69
+ 1. Read UniProtKB release 2026_03, every element with its own evidence codes: the 575,748 Swiss-Prot entries and 1,083,685 TrEMBL entries (1,033,251 of annotation score 4 or 5 or protein existence 1, streamed pinned to the release, and 50,434 that an experimental GOA line names), 42,036,351 statements; then GOA's 694,440,912 lines, of which the pool's accessions hold 28,164,146; InterPro's 904,908,849 match lines, which gave 39,403,909 protein-level and located InterPro and Pfam rows; Reactome's 223,594 pathway assignments; and the accessions' lineages, 13,158,683 taxonomy terms.
70
+ 2. Gave every statement the tier its evidence code earns: experimental (ECO:0000269 and its family, ECO:0007744, the GO experimental and high-throughput codes), signature (InterPro and Pfam matches), curated (ECO:0000250, 0000255 and 0000305; GO ISS, ISO, ISA, ISM, IC and TAS) or guard (every other code, kept but never a label or a negative); a keyword takes the tier of what implies it, and 1,799,118 keywords of the technical, coding-sequence-diversity, disease and developmental-stage categories were dropped.
71
+ 3. Pooled distinct sequences: dropped 30,918 fragments and 4,946 sequences over 2,046 residues, merged 142,405 accessions into the entry of an identical sequence, and kept the statements of the pooled accessions, 1,481,164 sequences with 118,715,542 assertions (14,911,139 experimental, 38,690,305 signature, 9,538,369 curated, 55,530,620 guard and 45,109 of codes that assert nothing).
72
+ 4. Clustered with MMseqs2 at 0.95, 0.9 and 0.8 identity (linclust) and 0.7, 0.6, 0.5, 0.4 and 0.3 (sensitivity 4, 5, 6, 7 and 7.5), 0.8 coverage of both sequences, each level clustering the representatives of the one before: 1,098,016 clusters at 0.95, 837,556 at 0.9, 590,815 at 0.8, 390,350 at 0.7, 302,677 at 0.6, 237,715 at 0.5, 190,909 at 0.4 and 158,689 at 0.3.
73
+ 5. Searched every sequence against every other once (MMseqs2 at sensitivity 5.7 and 300 candidates, e-value 1e-3 scaled to each split's residues, half of the query aligned) and split by cluster 85/5/8/2 (seed 42), holding out only clusters with an experimental assertion outside the organism aspect, balanced over the 11,982 experimental terms of 10 or more proteins, enforcing 0.5 identity; repair returned 162,452 clusters with a relative in another split to train.
74
+ 6. Checked the split: the twelve directed comparisons between the splits found no pair at 0.5 identity or more (the highest is 0.499), and the worst label-balance excess (a held-out split's distance from train's label shares over a random sample's of its size) is 2.11, on test, within the 2.5 the release states: the splits convention's allowance at any share, where test's own at 8 percent is 2.6.
75
+ 7. Kept in the vocabulary the 126,003 terms that 10 or more train proteins carry at a label tier, in eight fundamental aspects; wrote the views: whole_vocabulary (1,481,135 rows over 126,003 tokens), GO BP (125,864 rows over 9,050 terms), GO MF (106,843 rows over 3,288 terms), GO CC (149,430 rows over 1,664 terms), localization (343,774 rows over 2,015 terms) and EC (175,460 enzymes with one EC number, 11,965 with several left out); and scored the train prior and identity transfer from the split's own table on every view's validation and test rows, keeping each held-out protein's 50 nearest train proteins.
76
+
77
+ ![How protein_annotation_compendium_v1 was built](pipeline.svg)
78
+
79
+ <details><summary>Row counts at every step</summary>
80
+
81
+ | Step | Rows in | Rows out |
82
+ |---|---|---|
83
+ | the six line sources | 1,654,869,613 | 1,654,869,613 |
84
+ | GOA: lines of the pool's accessions (`other_database` removes 34,180,904; `isoform_or_form` removes 10,674; `accession_outside_pool` removes 632,085,188) | 1,654,869,613 | 988,592,847 |
85
+ | InterPro: matches of the pool's proteins (`interpro_other_protein` removes 885,846,109) | 988,592,847 | 102,746,738 |
86
+ | InterPro: protein-level and Pfam rows | 102,746,738 | 123,087,907 |
87
+ | Reactome: pathways of the pool's proteins (`reactome_isoform` removes 1,720; `reactome_other_protein` removes 99,504) | 123,087,907 | 122,986,683 |
88
+ | terms, hierarchy and keyword tiers | 122,986,683 | 122,986,683 |
89
+ | pooled accessions, kept keywords, GO terms (`swissprot_accession_not_in_pool` removes 418,439; `swissprot_dropped_keyword_category` removes 504,046; `trembl_accession_not_in_pool` removes 295,480; `trembl_dropped_keyword_category` removes 1,295,072; `goa_accession_not_in_pool` removes 752,080; `go_unresolved_or_other_aspect` removes 3,261; `interpro_accession_not_in_pool` removes 713,604; `reactome_accession_not_in_pool` removes 4,213; `taxonomy_accession_not_in_pool` removes 284,946) | 122,986,683 | 118,715,542 |
90
+ | union of assertions per sequence | 118,715,542 | 1,481,164 |
91
+ | cascade 0.95 to 0.3 and an all-pairs search | 1,481,164 | 1,481,164 |
92
+ | 85/5/8/2 by cluster | 1,481,164 | see splits |
93
+
94
+ </details>
95
+
96
+ Every source and filter is also listed as text in [pipeline.md](pipeline.md).
97
+
98
+ ![What the compendium holds, by aspect, tier and split](composition.png)
99
+
100
+ **What the compendium holds, by aspect, tier and split.** A, each fundamental aspect's protein-term pairs by their best evidence tier, with the total: domain and family 27.4M, 82 percent signatures; organism 11.8M, every one a lineage term; the other six 1.9M to 9.7M, of which 0.5 to 7.4 percent are experimental and most are guards. B, vocabulary terms per aspect (10 or more train proteins at a label tier), 126,003 in all, from 77,276 domain and family terms to 23 biophysical ones. C, proteins per split from Swiss-Prot and TrEMBL: train is 69 percent TrEMBL and the held-out splits 62 to 74 percent Swiss-Prot, since only clusters with an experimental assertion are held out. D, each held-out split's highest identity to a train protein in the split's table, cumulative: about a quarter have no hit, the median is 0.36, and none reaches the enforced 0.5 (dashed). E, the tier mix of each split's positive assertions: 12 percent experimental in train, 18 to 22 percent in the held-out splits. F, held-out AUROC of a logistic regression on metadata alone for whether a protein carries a label in each aspect: lineage and length (circles) from 0.59 to 0.87, the whole shortcut registry (squares) from 0.84 to 0.93, against chance at 0.5 and the gate's 0.55 for constructed negatives, which this release has none of; organism, which every protein holds, is left out.
101
+
102
+ ## Columns
103
+
104
+ The first columns are the ones to read. The ids and cluster columns after them are for joining and audit.
105
+
106
+ ### whole_vocabulary
107
+
108
+ | Column | What it holds |
109
+ |---|---|
110
+ | `name` | The protein's first Swiss-Prot accession, else its first TrEMBL one, as `UniProtKB:P77619`. |
111
+ | `seqs` | The protein sequence: a distinct sequence of at most 2,046 residues, never a fragment. |
112
+ | `tokens` | The vocabulary tokens of the terms the protein carries at a label tier, ascending; `tables/tokens.parquet` maps each to its term. |
113
+ | `terms` | The term of each token, in the same order, such as `av::go_bp::GO::GO:0006915`. |
114
+ | `tiers` | The best evidence tier of each term, in the same order: `experimental`, `signature`, or `curated` (in train and calibration only). |
115
+ | `masked` | Terms the protein holds only at a tier that neither scores nor counts as a negative here: a guard anywhere, a curated term in validation and test. |
116
+ | `observed` | The fundamental aspects the protein has a positive in; a term of another aspect is unknown for it, never a negative. |
117
+ | `protein_id` | SHA-256 of the sequence. Protify's pinned format requires it; it is an id, not something to read. |
118
+ | `cluster_id` | The protein's finest cluster (0.9 identity). Every member of a cluster sits in one split. |
119
+
120
+ ### go_bp
121
+
122
+ Also holds, with the same meaning as above: `name`, `seqs`, `masked`, `protein_id`, `cluster_id`.
123
+
124
+ | Column | What it holds |
125
+ |---|---|
126
+ | `positives` | The view's terms the protein carries at a label tier, with every ancestor along is_a and part_of that is a view term. |
127
+ | `negated` | View terms an experimental NOT assertion denies for the protein: explicit negatives. |
128
+
129
+ ### go_mf
130
+
131
+ Also holds, with the same meaning as above: `name`, `seqs`, `positives`, `masked`, `negated`, `protein_id`, `cluster_id`.
132
+
133
+ ### go_cc
134
+
135
+ Also holds, with the same meaning as above: `name`, `seqs`, `positives`, `masked`, `negated`, `protein_id`, `cluster_id`.
136
+
137
+ ### localization
138
+
139
+ Also holds, with the same meaning as above: `name`, `seqs`, `positives`, `masked`, `negated`, `protein_id`, `cluster_id`.
140
+
141
+ ### ec
142
+
143
+ Also holds, with the same meaning as above: `name`, `seqs`, `protein_id`, `cluster_id`.
144
+
145
+ | Column | What it holds |
146
+ |---|---|
147
+ | `level_1` | The protein's one EC number to its level 1, the rest dashed, such as `3.-.-.-`. |
148
+ | `level_2` | The protein's one EC number to its level 2, the rest dashed, such as `3.4.-.-`. |
149
+ | `level_3` | The protein's one EC number to its level 3, the rest dashed, such as `3.4.21.-`. |
150
+ | `level_4` | The protein's one EC number to its level 4, the rest dashed, such as `3.4.21.4`. |
151
+
152
+ ## Example rows
153
+
154
+ | name | seqs | terms |
155
+ |---|---|---|
156
+ | UniProtKB:P81809 | `ASPYAFGL` | av::keyword::UniProtKB-KW... |
157
+ | UniProtKB:A7Z3X4 | `MSEVLATYLLTDTTDSEKRAEQIAA...` | av::interpro::InterPro::I... |
158
+ | UniProtKB:A9MS91 | `MIKERRTELVEGFRHSVPYINTHRG...` | av::interpro::InterPro::I... |
159
+
160
+ ## What a model trained on it may say
161
+
162
+ It says how likely a protein is to carry each vocabulary term at the tier it was trained on. A low score does not show a term
163
+ absent, since a protein unobserved in an aspect has no negatives there; a signature term means InterPro's models match, not
164
+ that a function was measured; and an organism term names where the sequence was read, not a property of the protein.
165
+
166
+ ## Sources and licence
167
+
168
+ - UniProtKB/Swiss-Prot XML 2026_03: CC-BY-4.0
169
+ - UniProtKB/TrEMBL 2026_03 entries: CC-BY-4.0
170
+ - GOA UniProt, all species (2026-10-11): CC-BY-4.0
171
+ - InterPro protein2ipr (2026-08-31): CC0-1.0
172
+ - Reactome UniProt2Reactome (2026-06-19): CC0-1.0
173
+ - NCBI Taxonomy lineages of the entries: NCBI, no restrictions on use
174
+ - Ontologies and UniProt lists: CC-BY-4.0 or CC0-1.0
175
+ - 8 more sources were considered and set aside; each is listed with its reason in [pipeline.md](pipeline.md).
176
+
177
+ ## Files
178
+
179
+ | File | What it is |
180
+ |---|---|
181
+ | `{train,validation,test,calibration}.parquet` | the four splits of `whole_vocabulary` |
182
+ | `go_bp/{train,validation,test,calibration}.parquet` | the four splits of `go_bp` |
183
+ | `go_mf/{train,validation,test,calibration}.parquet` | the four splits of `go_mf` |
184
+ | `go_cc/{train,validation,test,calibration}.parquet` | the four splits of `go_cc` |
185
+ | `localization/{train,validation,test,calibration}.parquet` | the four splits of `localization` |
186
+ | `ec/{train,validation,test,calibration}.parquet` | the four splits of `ec` |
187
+ | `pipeline.md` | the pipeline as text, for agents |
188
+ | `card_facts.json` | the numbers this card states, written by the exporter from the build's receipts |
189
+ | `receipts/` | the build's records: `build.json` (the sources by digest, every stage's counts, time and peak memory), `split.json` (clusters, strata, repair and the twelve directed searches), `survey.json` (the identity survey that chose the enforced level), `audit.json` (each split's tier mix and what metadata alone scores), `baselines.json` (each view's prior and identity transfer on validation and test) and `acceptance.json` (each acceptance check) |
190
+ | `tables/assertions.parquet` | every pooled assertion: protein, term, aspect, evidence code, tier, source, reference and residue positions |
191
+ | `tables/clusters.parquet` | each protein's cluster at 0.95, 0.9, 0.8, 0.7, 0.6, 0.5, 0.4 and 0.3 identity |
192
+ | `tables/concepts.parquet` | the names and definitions of the ontology concepts the terms name |
193
+ | `tables/hierarchy.parquet` | the typed edges between concepts: GO, EC, taxonomy, InterPro, subcellular location, keywords, ChEBI and PSI-MOD |
194
+ | `tables/measurement_texts.parquet` | the biophysicochemical statements of the entries, raw, for the measurement readers |
195
+ | `tables/neighbours.parquet` | each held-out protein's 50 nearest train proteins in the split's own search, by identity, with the e-value: the identity baseline's input |
196
+ | `tables/proteins.parquet` | every pooled protein: sequence, accessions, review state, evidence level, annotation score, lineage and study metadata |
197
+ | `tables/split.parquet` | each protein's split, whether it could be held out, and a held-out protein's stratum and identity to train |
198
+ | `tables/tokens.parquet` | each vocabulary term's token |
199
+ | `tables/vocabulary.parquet` | every term with its support per tier and split, and whether it is in the vocabulary (10 train proteins at a label tier) |
200
+ | `composition.png` | a figure of the build, shown above |
calibration.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ac53f8d634721583b8006a51f8de2856599fc3e497def34eae99273f75a7755c
3
+ size 3463600
card_facts.json ADDED
@@ -0,0 +1,112 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "acceptance": "5 of 5 acceptance checks passed, among them 12 directed homology searches with none at 0.5 identity or more",
3
+ "cluster_label": "Clusters (0.9 identity)",
4
+ "column_notes": {
5
+ "level_1": "The protein's one EC number to its level 1, the rest dashed, such as `3.-.-.-`.",
6
+ "level_2": "The protein's one EC number to its level 2, the rest dashed, such as `3.4.-.-`.",
7
+ "level_3": "The protein's one EC number to its level 3, the rest dashed, such as `3.4.21.-`.",
8
+ "level_4": "The protein's one EC number to its level 4, the rest dashed, such as `3.4.21.4`.",
9
+ "masked": "Terms the protein holds only at a tier that neither scores nor counts as a negative here: a guard anywhere, a curated term in validation and test.",
10
+ "name": "The protein's first Swiss-Prot accession, else its first TrEMBL one, as `UniProtKB:P77619`.",
11
+ "negated": "View terms an experimental NOT assertion denies for the protein: explicit negatives.",
12
+ "observed": "The fundamental aspects the protein has a positive in; a term of another aspect is unknown for it, never a negative.",
13
+ "positives": "The view's terms the protein carries at a label tier, with every ancestor along is_a and part_of that is a view term.",
14
+ "seqs": "The protein sequence: a distinct sequence of at most 2,046 residues, never a fragment.",
15
+ "terms": "The term of each token, in the same order, such as `av::go_bp::GO::GO:0006915`.",
16
+ "tiers": "The best evidence tier of each term, in the same order: `experimental`, `signature`, or `curated` (in train and calibration only).",
17
+ "tokens": "The vocabulary tokens of the terms the protein carries at a label tier, ascending; `tables/tokens.parquet` maps each to its term."
18
+ },
19
+ "configs": [
20
+ {
21
+ "directory": "",
22
+ "name": "whole_vocabulary"
23
+ },
24
+ {
25
+ "directory": "go_bp/",
26
+ "name": "go_bp"
27
+ },
28
+ {
29
+ "directory": "go_mf/",
30
+ "name": "go_mf"
31
+ },
32
+ {
33
+ "directory": "go_cc/",
34
+ "name": "go_cc"
35
+ },
36
+ {
37
+ "directory": "localization/",
38
+ "name": "localization"
39
+ },
40
+ {
41
+ "directory": "ec/",
42
+ "name": "ec"
43
+ }
44
+ ],
45
+ "criteria": null,
46
+ "dataset_id": "protein_annotation_compendium_v1",
47
+ "endpoint": "protein_annotation",
48
+ "example_columns": [
49
+ "name",
50
+ "seqs",
51
+ "terms"
52
+ ],
53
+ "file_notes": {
54
+ "receipts/": "the build's records: `build.json` (the sources by digest, every stage's counts, time and peak memory), `split.json` (clusters, strata, repair and the twelve directed searches), `survey.json` (the identity survey that chose the enforced level), `audit.json` (each split's tier mix and what metadata alone scores), `baselines.json` (each view's prior and identity transfer on validation and test) and `acceptance.json` (each acceptance check)",
55
+ "tables/assertions.parquet": "every pooled assertion: protein, term, aspect, evidence code, tier, source, reference and residue positions",
56
+ "tables/clusters.parquet": "each protein's cluster at 0.95, 0.9, 0.8, 0.7, 0.6, 0.5, 0.4 and 0.3 identity",
57
+ "tables/concepts.parquet": "the names and definitions of the ontology concepts the terms name",
58
+ "tables/hierarchy.parquet": "the typed edges between concepts: GO, EC, taxonomy, InterPro, subcellular location, keywords, ChEBI and PSI-MOD",
59
+ "tables/measurement_texts.parquet": "the biophysicochemical statements of the entries, raw, for the measurement readers",
60
+ "tables/neighbours.parquet": "each held-out protein's 50 nearest train proteins in the split's own search, by identity, with the e-value: the identity baseline's input",
61
+ "tables/proteins.parquet": "every pooled protein: sequence, accessions, review state, evidence level, annotation score, lineage and study metadata",
62
+ "tables/split.parquet": "each protein's split, whether it could be held out, and a held-out protein's stratum and identity to train",
63
+ "tables/tokens.parquet": "each vocabulary term's token",
64
+ "tables/vocabulary.parquet": "every term with its support per tier and split, and whether it is in the vocabulary (10 train proteins at a label tier)"
65
+ },
66
+ "homology": {
67
+ "audit_threshold": 0.5,
68
+ "coverage": 0.8,
69
+ "enforced": 0.5,
70
+ "far": 18767,
71
+ "held_out": 57172,
72
+ "largest": 0.499,
73
+ "levels": [
74
+ 0.9,
75
+ 0.7,
76
+ 0.5,
77
+ 0.3
78
+ ],
79
+ "near_duplicates": 0,
80
+ "rationale": "CAMP, which reads this split, may enforce 0.5 at most, and no level up to it is well formed, so the survey of the clusters recommends 0.5, the finest such level: there 886 of the 6,066 labels spread over 20 or more units at 0.9 identity fall to fewer (14.6 percent, over the survey's 10), as their proteins sit in a few large families; 0.7 is the lowest well-formed level. Repair returned to train every held-out cluster with a relative in another split at 0.5 identity, since the clusters join 0.8 of both sequences and the audit counts half of one.",
81
+ "searches": 12,
82
+ "unit": "sequence"
83
+ },
84
+ "labels": [],
85
+ "splits": {
86
+ "calibration": {
87
+ "clusters": 5167,
88
+ "proteins": 6830,
89
+ "rows": 6830
90
+ },
91
+ "test": {
92
+ "clusters": 21515,
93
+ "proteins": 31661,
94
+ "rows": 31661
95
+ },
96
+ "train": {
97
+ "clusters": 798403,
98
+ "proteins": 1423964,
99
+ "rows": 1423964
100
+ },
101
+ "validation": {
102
+ "clusters": 12446,
103
+ "proteins": 18680,
104
+ "rows": 18680
105
+ }
106
+ },
107
+ "target": "tokens",
108
+ "task": "prediction of a protein's annotation terms from its sequence over one vocabulary in eight fundamental aspects (Oracle One, the translator), retrieval between a sequence and its annotation (CAMP), and per-aspect views for GO, localization and EC",
109
+ "task_type": "multilabel",
110
+ "unit_label": "Proteins",
111
+ "units": null
112
+ }
composition.png ADDED

Git LFS Details

  • SHA256: 5c5415139bd9c58d77697db0970a6d2064151dd546cc9ac55485e42fca4fe509
  • Pointer size: 131 Bytes
  • Size of remote file: 528 kB
ec/calibration.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:20eaa12bb21cfcd6381255e7925a607287995aa1dc7ff2aa1615520667b48548
3
+ size 590628
ec/test.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7dde4fade353ee837a977eae79473660bda564ec89feb16a992de65951e095cd
3
+ size 531002
ec/train.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7a0f71637dd0b752425e88aef46a7b1029294ada56ec8e30850a1ce6781419ac
3
+ size 52030767
ec/validation.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dc83d40d094f82dde4539229ecec1f2ef9dbc84ce383f835f10b993617e07426
3
+ size 334870
go_bp/calibration.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4260287f039a512275bb80c85c5129d9ea8b6e59e0b101210da4d178b2e18040
3
+ size 642195
go_bp/test.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1d2cecdf0ce43c631e7bd9714c87b34e1d9d191eda709d87c1ee00aa0c307e45
3
+ size 2240172
go_bp/train.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0d357b6cf96d635062a2b42af5acb66a95e9f12b57a1a41a70572bb35dd048fa
3
+ size 47757238
go_bp/validation.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3cdb574b81c9e554e4514e609e6a77ce6db394d76802d47df56e37a1bb72b833
3
+ size 1395892
go_cc/calibration.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cc8c6636888a81f196d66eaee7480163d7e4b6a52f975d447bc30b3bc4c687a5
3
+ size 703462
go_cc/test.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:11d3075c74ba5c60fbaa2c55c67db24369272e490e2f9c7f8fe69f4245bb4f13
3
+ size 2204116
go_cc/train.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e1f24b5bd6cc71466fc61e936cc9bf6c4a0ca8dee9a630c1ea940363b05934fd
3
+ size 48984981
go_cc/validation.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:74e8a9513189675fbcd5bcb7ac5170c5b1b4819bc154b52a5b5277cb20d8710c
3
+ size 1444374
go_mf/calibration.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5ef698cb2ce3b23793c2afe9b542f9c089e114b34326b841c41ef1e573d44341
3
+ size 505851
go_mf/test.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0c8235a82ff229c00bcca25430341489751b4cc7f45d23e4c98b376b1b0bbd33
3
+ size 1466604
go_mf/train.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0c9e6b639c6e6a08989c5968f926764371f027a49af0dae871cc3c55a9366f10
3
+ size 36720319
go_mf/validation.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cd87e2df3d614d77f2d03da667dbe6242b5f6ed46222fd6f3d7ee6dc4c69df10
3
+ size 932207
localization/calibration.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4ecfb132d46527f9ea9774265f1d973c38d07300080be87a48c221d5624c8d15
3
+ size 1313445
localization/test.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:86f74e7431d600488da1d12479db98225984795ec4cea4a3299918157672c854
3
+ size 2360909
localization/train.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:819be918784d53e3cdf12030d5d7eb08f099604971e35c221bde1eeecb4ade03
3
+ size 103325398
localization/validation.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:41d8db2ffd4016f4e4cd542f7c12474d6bc3494a272c3ee3089a06a97b556e92
3
+ size 1556293
pipeline.md ADDED
@@ -0,0 +1,56 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <!-- ws-dataset-map spec_sha256=c92cb58ce91316d16794372e86bc0cb5aab8df75117d90a2cad80503ad1f8627 -->
2
+ # How protein_annotation_compendium_v1 was built
3
+
4
+ This file is generated from `pipeline.yaml` by `ws dataset map render` and lists what the map in `pipeline.svg` or `pipeline.gif` shows. Edit the spec and render again; never edit this file.
5
+
6
+ - Provenance: **recorded**, every count comes from the build's own record, and the counts close from the sources to the splits.
7
+ - Record: `docs/decisions/representation/annotation_vocabulary/2026-10-20_annotation_compendium.md`
8
+ - Source register: section `Annotation compendium` of `docs/guides/dataset_sources.md`
9
+
10
+ ## Sources
11
+
12
+ | Id | Source | Role | Terms | Rows | SHA-256 |
13
+ |---|---|---|---|---|---|
14
+ | `swissprot` | UniProtKB/Swiss-Prot XML 2026_03 | rows: rows come from it | CC-BY-4.0 | 11,181,565 | |
15
+ | `trembl` | UniProtKB/TrEMBL 2026_03 entries | rows: rows come from it | CC-BY-4.0 | 30,854,786 | |
16
+ | `goa` | GOA UniProt, all species (2026-10-11) | rows: rows come from it | CC-BY-4.0 | 694,440,912 | |
17
+ | `interpro` | InterPro protein2ipr (2026-08-31) | rows: rows come from it | CC0-1.0 | 904,908,849 | |
18
+ | `reactome` | Reactome UniProt2Reactome (2026-06-19) | rows: rows come from it | CC0-1.0 | 324,818 | |
19
+ | `taxonomy` | NCBI Taxonomy lineages of the entries | rows: rows come from it | NCBI, no restrictions on use | 13,158,683 | |
20
+ | `ontologies` | Ontologies and UniProt lists | reference: read to annotate or filter rows | CC-BY-4.0 or CC0-1.0 | | |
21
+ | `kegg` | KEGG pathways | excluded: set aside, not used | | | |
22
+ | `iptmnet` | iPTMnet, PhosphoSitePlus | excluded: set aside, not used | | | |
23
+ | `disease` | Disease terms | excluded: set aside, not used | | | |
24
+ | `family_trees` | GO IBA, IBD, IKR, IRD (guards, not labels) | excluded: set aside, not used | | | |
25
+ | `rules` | IEA, UniRule, ARBA (guards, not labels) | excluded: set aside, not used | | | |
26
+ | `glygen` | GlyGen | lead: set aside, not yet examined | | | |
27
+ | `measurements` | BRENDA, SABIO-RK, Meltome | lead: set aside, not yet examined | | | |
28
+ | `families` | CAZy, MEROPS, TCDB (licence pending) | lead: set aside, not yet examined | | | |
29
+
30
+ ## Steps
31
+
32
+ | # | Stage | Step | Kind | Line | Rows in | Rows out | Detail |
33
+ |---|---|---|---|---|---|---|---|
34
+ | 1 | Read the line sources | the six line sources | merge | rows | 1,654,869,613 | 1,654,869,613 | joins lines `swissprot`, `trembl`, `goa`, `interpro`, `reactome`, `taxonomy` into one rows line |
35
+ | 2 | Read the line sources | GOA: lines of the pool's accessions | filter | rows | 1,654,869,613 | 988,592,847 | `other_database` removes 34,180,904; `isoform_or_form` removes 10,674; `accession_outside_pool` removes 632,085,188 |
36
+ | 3 | Read the line sources | InterPro: matches of the pool's proteins | filter | rows | 988,592,847 | 102,746,738 | `interpro_other_protein` removes 885,846,109 |
37
+ | 4 | Read the line sources | InterPro: protein-level and Pfam rows | add | rows | 102,746,738 | 123,087,907 | `protein_level_and_pfam_rows` adds 20,341,169 |
38
+ | 5 | Read the line sources | Reactome: pathways of the pool's proteins | filter | rows | 123,087,907 | 122,986,683 | `reactome_isoform` removes 1,720; `reactome_other_protein` removes 99,504 |
39
+ | 6 | Pool the accessions | terms, hierarchy and keyword tiers | join | rows | 122,986,683 | 122,986,683 | reads reference `ontologies` |
40
+ | 7 | Pool the accessions | pooled accessions, kept keywords, GO terms | filter | rows | 122,986,683 | 118,715,542 | `swissprot_accession_not_in_pool` removes 418,439; `swissprot_dropped_keyword_category` removes 504,046; `trembl_accession_not_in_pool` removes 295,480; `trembl_dropped_keyword_category` removes 1,295,072; `goa_accession_not_in_pool` removes 752,080; `go_unresolved_or_other_aspect` removes 3,261; `interpro_accession_not_in_pool` removes 713,604; `reactome_accession_not_in_pool` removes 4,213; `taxonomy_accession_not_in_pool` removes 284,946 |
41
+ | 8 | Pool the accessions | union of assertions per sequence | collapse | rows | 118,715,542 | 1,481,164 | computes union of assertions per sequence; rows_after is 1,481,164 |
42
+ | 9 | Cluster and split | cascade 0.95 to 0.3 and an all-pairs search | process | rows | 1,481,164 | 1,481,164 | |
43
+ | 10 | Cluster and split | 85/5/8/2 by cluster | split | rows | 1,481,164 | | train 1,423,992, validation 18,680, test 31,662, calibration 6,830 |
44
+
45
+ ## Splits
46
+
47
+ | Split | Rows | Share |
48
+ |---|---|---|
49
+ | train | 1,423,992 | 96.1% |
50
+ | validation | 18,680 | 1.3% |
51
+ | test | 31,662 | 2.1% |
52
+ | calibration | 6,830 | 0.5% |
53
+
54
+ ## Closure
55
+
56
+ 1,654,869,613 rows are read from 6 rows sources, filters remove 1,556,495,240, steps add 20,341,169, collapsing removes 117,234,378, and 1,481,164 reach the split. Every source and filter the record names is on the map. Sources marked set aside were examined and not used; they are listed above and drawn grey in the legend.
pipeline.svg ADDED
receipts/acceptance.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "checks": [
3
+ {
4
+ "detail": "12 directed searches at 0.5 identity, the largest cross-split identity 0.499",
5
+ "name": "homology_audit",
6
+ "passed": true
7
+ },
8
+ {
9
+ "detail": "worst excess 2.11 over 11,982 terms, 2.5 at most",
10
+ "name": "label_balance",
11
+ "passed": true
12
+ },
13
+ {
14
+ "detail": "validation 18,680, test 31,662, calibration 6,830 proteins, 1,000 at least",
15
+ "name": "held_out_size",
16
+ "passed": true
17
+ },
18
+ {
19
+ "detail": "1,481,164 split rows for 1,481,164 pooled proteins",
20
+ "name": "split_covers_pool",
21
+ "passed": true
22
+ },
23
+ {
24
+ "detail": "0 rows in another split than the split table's; every view row written: {'whole_vocabulary': True, 'go_bp': True, 'go_mf': True, 'go_cc': True, 'localization': True, 'ec': True}",
25
+ "name": "configs_hold_their_views",
26
+ "passed": true
27
+ }
28
+ ],
29
+ "criteria_passed": true
30
+ }
receipts/audit.json ADDED
@@ -0,0 +1,74 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "observation_baselines": {
3
+ "biophysical_assembly": {
4
+ "gated_auroc": 0.6464360040816528,
5
+ "registry_auroc": 0.8443156546622663,
6
+ "share": 0.07122911439921575
7
+ },
8
+ "domain_family": {
9
+ "gated_auroc": 0.8712819932986469,
10
+ "registry_auroc": 0.9297241782944408,
11
+ "share": 0.9858320888166334
12
+ },
13
+ "enzyme": {
14
+ "gated_auroc": 0.7238169038603505,
15
+ "registry_auroc": 0.8999007312062784,
16
+ "share": 0.1777284622094515
17
+ },
18
+ "localization": {
19
+ "gated_auroc": 0.6058733722325523,
20
+ "registry_auroc": 0.8661524323902198,
21
+ "share": 0.2477875508721519
22
+ },
23
+ "molecular_function": {
24
+ "gated_auroc": 0.6434872060571002,
25
+ "registry_auroc": 0.9144806587598716,
26
+ "share": 0.2837038977452868
27
+ },
28
+ "organism": {
29
+ "gated_auroc": 0.9988940983259165,
30
+ "registry_auroc": 0.9999304672600127,
31
+ "share": 0.9999723190679762
32
+ },
33
+ "process_pathway": {
34
+ "gated_auroc": 0.6034572893836972,
35
+ "registry_auroc": 0.839185390883925,
36
+ "share": 0.25294835683286926
37
+ },
38
+ "ptm_processing": {
39
+ "gated_auroc": 0.5939293699254157,
40
+ "registry_auroc": 0.8872901784516231,
41
+ "share": 0.08491159655514177
42
+ }
43
+ },
44
+ "tier_mix": {
45
+ "calibration": {
46
+ "curated": 79078,
47
+ "experimental": 83546,
48
+ "guard": 96035,
49
+ "signature": 122387,
50
+ "skip": 716
51
+ },
52
+ "test": {
53
+ "curated": 373755,
54
+ "experimental": 388958,
55
+ "guard": 698601,
56
+ "signature": 596387,
57
+ "skip": 2055
58
+ },
59
+ "train": {
60
+ "curated": 8885291,
61
+ "experimental": 14208496,
62
+ "guard": 54269178,
63
+ "signature": 37628530,
64
+ "skip": 40864
65
+ },
66
+ "validation": {
67
+ "curated": 197909,
68
+ "experimental": 226177,
69
+ "guard": 465144,
70
+ "signature": 343001,
71
+ "skip": 1474
72
+ }
73
+ }
74
+ }
receipts/baselines.json ADDED
@@ -0,0 +1,320 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "table": {
3
+ "evalue_max": 0.001,
4
+ "hits": 308156121,
5
+ "reach": 400.0,
6
+ "residues": 785205992,
7
+ "sequences": 1481164,
8
+ "source": "all_pairs"
9
+ },
10
+ "views": {
11
+ "ec": {
12
+ "test": {
13
+ "level_1": {
14
+ "classes": 7,
15
+ "identity": {
16
+ "accuracy": 0.8334,
17
+ "macro_f1": 0.8874
18
+ },
19
+ "prior": {
20
+ "accuracy": 0.3211,
21
+ "macro_f1": 0.0694
22
+ }
23
+ },
24
+ "level_2": {
25
+ "classes": 57,
26
+ "identity": {
27
+ "accuracy": 0.7973,
28
+ "macro_f1": 0.7359
29
+ },
30
+ "prior": {
31
+ "accuracy": 0.095,
32
+ "macro_f1": 0.003
33
+ }
34
+ },
35
+ "level_3": {
36
+ "classes": 173,
37
+ "identity": {
38
+ "accuracy": 0.7699,
39
+ "macro_f1": 0.6851
40
+ },
41
+ "prior": {
42
+ "accuracy": 0.0033,
43
+ "macro_f1": 0.0
44
+ }
45
+ },
46
+ "level_4": {
47
+ "classes": 1045,
48
+ "identity": {
49
+ "accuracy": 0.4114,
50
+ "macro_f1": 0.3672
51
+ },
52
+ "prior": {
53
+ "accuracy": 0.0027,
54
+ "macro_f1": 0.0
55
+ }
56
+ },
57
+ "rows": 1495,
58
+ "with_train_hit": 1313
59
+ },
60
+ "validation": {
61
+ "level_1": {
62
+ "classes": 7,
63
+ "identity": {
64
+ "accuracy": 0.813,
65
+ "macro_f1": 0.84
66
+ },
67
+ "prior": {
68
+ "accuracy": 0.2895,
69
+ "macro_f1": 0.0641
70
+ }
71
+ },
72
+ "level_2": {
73
+ "classes": 56,
74
+ "identity": {
75
+ "accuracy": 0.7901,
76
+ "macro_f1": 0.8036
77
+ },
78
+ "prior": {
79
+ "accuracy": 0.0856,
80
+ "macro_f1": 0.0028
81
+ }
82
+ },
83
+ "level_3": {
84
+ "classes": 147,
85
+ "identity": {
86
+ "accuracy": 0.7515,
87
+ "macro_f1": 0.7142
88
+ },
89
+ "prior": {
90
+ "accuracy": 0.0012,
91
+ "macro_f1": 0.0
92
+ }
93
+ },
94
+ "level_4": {
95
+ "classes": 620,
96
+ "identity": {
97
+ "accuracy": 0.4499,
98
+ "macro_f1": 0.4177
99
+ },
100
+ "prior": {
101
+ "accuracy": 0.0024,
102
+ "macro_f1": 0.0
103
+ }
104
+ },
105
+ "rows": 829,
106
+ "with_train_hit": 716
107
+ }
108
+ },
109
+ "go_bp": {
110
+ "test": {
111
+ "identity": {
112
+ "entries": 233308,
113
+ "fmax": 0.2924,
114
+ "macro_ap": 0.1216,
115
+ "micro_ap": 0.0812,
116
+ "scored_terms": 1085,
117
+ "threshold": 0.3,
118
+ "with_train_hit": 3762
119
+ },
120
+ "prior": {
121
+ "entries": 2252868,
122
+ "fmax": 0.111,
123
+ "macro_ap": 0.0084,
124
+ "micro_ap": 0.0464,
125
+ "scored_terms": 1085,
126
+ "threshold": 0.1
127
+ },
128
+ "rows": 5324
129
+ },
130
+ "validation": {
131
+ "identity": {
132
+ "entries": 157250,
133
+ "fmax": 0.3026,
134
+ "macro_ap": 0.1334,
135
+ "micro_ap": 0.0952,
136
+ "scored_terms": 702,
137
+ "threshold": 0.29,
138
+ "with_train_hit": 2319
139
+ },
140
+ "prior": {
141
+ "entries": 1372731,
142
+ "fmax": 0.1115,
143
+ "macro_ap": 0.0116,
144
+ "micro_ap": 0.0507,
145
+ "scored_terms": 702,
146
+ "threshold": 0.1
147
+ },
148
+ "rows": 3244
149
+ }
150
+ },
151
+ "go_cc": {
152
+ "test": {
153
+ "identity": {
154
+ "entries": 75020,
155
+ "fmax": 0.4665,
156
+ "macro_ap": 0.1609,
157
+ "micro_ap": 0.2118,
158
+ "scored_terms": 274,
159
+ "threshold": 0.22,
160
+ "with_train_hit": 3904
161
+ },
162
+ "prior": {
163
+ "entries": 720765,
164
+ "fmax": 0.4547,
165
+ "macro_ap": 0.0214,
166
+ "micro_ap": 0.3672,
167
+ "scored_terms": 274,
168
+ "threshold": 0.26
169
+ },
170
+ "rows": 6042
171
+ },
172
+ "validation": {
173
+ "identity": {
174
+ "entries": 48840,
175
+ "fmax": 0.4622,
176
+ "macro_ap": 0.1644,
177
+ "micro_ap": 0.2279,
178
+ "scored_terms": 197,
179
+ "threshold": 0.21,
180
+ "with_train_hit": 2537
181
+ },
182
+ "prior": {
183
+ "entries": 464373,
184
+ "fmax": 0.4539,
185
+ "macro_ap": 0.0285,
186
+ "micro_ap": 0.3741,
187
+ "scored_terms": 197,
188
+ "threshold": 0.26
189
+ },
190
+ "rows": 3891
191
+ }
192
+ },
193
+ "go_mf": {
194
+ "test": {
195
+ "identity": {
196
+ "entries": 34640,
197
+ "fmax": 0.5684,
198
+ "macro_ap": 0.4543,
199
+ "micro_ap": 0.4461,
200
+ "scored_terms": 298,
201
+ "threshold": 0.24,
202
+ "with_train_hit": 2875
203
+ },
204
+ "prior": {
205
+ "entries": 504090,
206
+ "fmax": 0.2308,
207
+ "macro_ap": 0.0152,
208
+ "micro_ap": 0.1564,
209
+ "scored_terms": 298,
210
+ "threshold": 0.1
211
+ },
212
+ "rows": 3842
213
+ },
214
+ "validation": {
215
+ "identity": {
216
+ "entries": 21835,
217
+ "fmax": 0.5863,
218
+ "macro_ap": 0.4427,
219
+ "micro_ap": 0.4454,
220
+ "scored_terms": 212,
221
+ "threshold": 0.27,
222
+ "with_train_hit": 1790
223
+ },
224
+ "prior": {
225
+ "entries": 307296,
226
+ "fmax": 0.2368,
227
+ "macro_ap": 0.0204,
228
+ "micro_ap": 0.1703,
229
+ "scored_terms": 212,
230
+ "threshold": 0.1
231
+ },
232
+ "rows": 2343
233
+ }
234
+ },
235
+ "localization": {
236
+ "test": {
237
+ "identity": {
238
+ "entries": 89819,
239
+ "fmax": 0.4518,
240
+ "macro_ap": 0.1604,
241
+ "micro_ap": 0.208,
242
+ "scored_terms": 348,
243
+ "threshold": 0.19,
244
+ "with_train_hit": 4329
245
+ },
246
+ "prior": {
247
+ "entries": 531745,
248
+ "fmax": 0.3612,
249
+ "macro_ap": 0.0192,
250
+ "micro_ap": 0.2903,
251
+ "scored_terms": 348,
252
+ "threshold": 0.2
253
+ },
254
+ "rows": 6421
255
+ },
256
+ "validation": {
257
+ "identity": {
258
+ "entries": 60146,
259
+ "fmax": 0.4456,
260
+ "macro_ap": 0.1703,
261
+ "micro_ap": 0.2199,
262
+ "scored_terms": 254,
263
+ "threshold": 0.19,
264
+ "with_train_hit": 2819
265
+ },
266
+ "prior": {
267
+ "entries": 345273,
268
+ "fmax": 0.3575,
269
+ "macro_ap": 0.0249,
270
+ "micro_ap": 0.2935,
271
+ "scored_terms": 254,
272
+ "threshold": 0.16
273
+ },
274
+ "rows": 4167
275
+ }
276
+ },
277
+ "whole_vocabulary": {
278
+ "test": {
279
+ "identity": {
280
+ "entries": 1676519,
281
+ "fmax": 0.3908,
282
+ "macro_ap": 0.5973,
283
+ "micro_ap": 0.2598,
284
+ "scored_terms": 7747,
285
+ "threshold": 0.32,
286
+ "with_train_hit": 24086
287
+ },
288
+ "prior": {
289
+ "entries": 9072743,
290
+ "fmax": 0.1542,
291
+ "macro_ap": 0.0034,
292
+ "micro_ap": 0.0621,
293
+ "scored_terms": 7747,
294
+ "threshold": 0.1
295
+ },
296
+ "rows": 31661
297
+ },
298
+ "validation": {
299
+ "identity": {
300
+ "entries": 971969,
301
+ "fmax": 0.3963,
302
+ "macro_ap": 0.6192,
303
+ "micro_ap": 0.2669,
304
+ "scored_terms": 4720,
305
+ "threshold": 0.32,
306
+ "with_train_hit": 13450
307
+ },
308
+ "prior": {
309
+ "entries": 5612459,
310
+ "fmax": 0.1427,
311
+ "macro_ap": 0.0051,
312
+ "micro_ap": 0.0619,
313
+ "scored_terms": 4720,
314
+ "threshold": 0.1
315
+ },
316
+ "rows": 18680
317
+ }
318
+ }
319
+ }
320
+ }
receipts/build.json ADDED
The diff for this file is too large to render. See raw diff
 
receipts/split.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c338484e2fde7e252f563fb3b6b9dd3f2dd3007e7010cce07dda4eb72cf90aac
3
+ size 75097202
receipts/survey.json ADDED
@@ -0,0 +1,158 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "ceiling": 0.5,
3
+ "clusters_per_split": 2753,
4
+ "finest_clusters": 137662,
5
+ "labels": 6066,
6
+ "levels": [
7
+ {
8
+ "failures": [
9
+ "labels_lost"
10
+ ],
11
+ "labels_carried": 4824,
12
+ "labels_lost": 1242,
13
+ "largest_unit": 187,
14
+ "level": 0.3,
15
+ "locked_share": 0.0,
16
+ "lost_labels": [
17
+ "av::cofactor::ChEBI::CHEBI:18408",
18
+ "av::cofactor::ChEBI::CHEBI:60539",
19
+ "av::ec::EC::1.1.1.1",
20
+ "av::ec::EC::1.1.1.2",
21
+ "av::ec::EC::1.1.1.62",
22
+ "av::ec::EC::1.11.1.24",
23
+ "av::ec::EC::1.14.14.1",
24
+ "av::ec::EC::1.14.99.54",
25
+ "av::ec::EC::1.14.99.56",
26
+ "av::ec::EC::1.15.1.1",
27
+ "av::ec::EC::1.16.3.1",
28
+ "av::ec::EC::1.2.1.3",
29
+ "av::ec::EC::1.4.3.21",
30
+ "av::ec::EC::1.6.2.4",
31
+ "av::ec::EC::2.1.1.37",
32
+ "av::ec::EC::2.3.1.199",
33
+ "av::ec::EC::2.3.1.20",
34
+ "av::ec::EC::2.3.1.23",
35
+ "av::ec::EC::2.3.1.286",
36
+ "av::ec::EC::2.3.1.31",
37
+ "av::ec::EC::2.3.1.46",
38
+ "av::ec::EC::2.3.1.51",
39
+ "av::ec::EC::2.3.2.23",
40
+ "av::ec::EC::2.3.2.26",
41
+ "av::ec::EC::2.4.1.41"
42
+ ],
43
+ "lost_share": 0.20474777448071216,
44
+ "singleton_units": 47204,
45
+ "units": 67767,
46
+ "units_per_split": 1355.221854978135,
47
+ "well_formed": false
48
+ },
49
+ {
50
+ "failures": [
51
+ "labels_lost"
52
+ ],
53
+ "labels_carried": 5180,
54
+ "labels_lost": 886,
55
+ "largest_unit": 93,
56
+ "level": 0.5,
57
+ "locked_share": 0.0,
58
+ "lost_labels": [
59
+ "av::cofactor::ChEBI::CHEBI:60539",
60
+ "av::ec::EC::1.1.1.1",
61
+ "av::ec::EC::1.1.1.2",
62
+ "av::ec::EC::1.1.1.62",
63
+ "av::ec::EC::1.11.1.24",
64
+ "av::ec::EC::1.14.14.1",
65
+ "av::ec::EC::1.15.1.1",
66
+ "av::ec::EC::1.16.3.1",
67
+ "av::ec::EC::1.2.1.3",
68
+ "av::ec::EC::1.4.3.21",
69
+ "av::ec::EC::1.6.2.4",
70
+ "av::ec::EC::2.1.1.37",
71
+ "av::ec::EC::2.3.1.199",
72
+ "av::ec::EC::2.3.1.23",
73
+ "av::ec::EC::2.3.1.286",
74
+ "av::ec::EC::2.3.1.46",
75
+ "av::ec::EC::2.3.2.26",
76
+ "av::ec::EC::2.5.1.47",
77
+ "av::ec::EC::2.6.1.1",
78
+ "av::ec::EC::2.7.1.107",
79
+ "av::ec::EC::2.7.1.11",
80
+ "av::ec::EC::2.7.11.22",
81
+ "av::ec::EC::2.7.11.24",
82
+ "av::ec::EC::2.7.4.6",
83
+ "av::ec::EC::2.7.7.19"
84
+ ],
85
+ "lost_share": 0.14606000659413124,
86
+ "singleton_units": 69078,
87
+ "units": 89994,
88
+ "units_per_split": 1799.7231044151617,
89
+ "well_formed": false
90
+ },
91
+ {
92
+ "failures": [],
93
+ "labels_carried": 5544,
94
+ "labels_lost": 522,
95
+ "largest_unit": 77,
96
+ "level": 0.7,
97
+ "locked_share": 0.0,
98
+ "lost_labels": [
99
+ "av::ec::EC::1.1.1.2",
100
+ "av::ec::EC::1.1.1.62",
101
+ "av::ec::EC::1.2.1.3",
102
+ "av::ec::EC::1.4.3.21",
103
+ "av::ec::EC::1.6.2.4",
104
+ "av::ec::EC::2.1.1.37",
105
+ "av::ec::EC::2.3.1.199",
106
+ "av::ec::EC::2.3.1.23",
107
+ "av::ec::EC::2.3.1.286",
108
+ "av::ec::EC::2.5.1.47",
109
+ "av::ec::EC::2.7.4.6",
110
+ "av::ec::EC::3.1.3.2",
111
+ "av::ec::EC::3.1.3.3",
112
+ "av::ec::EC::3.1.3.5",
113
+ "av::ec::EC::3.1.4.17",
114
+ "av::ec::EC::3.1.4.53",
115
+ "av::ec::EC::3.2.1.1",
116
+ "av::ec::EC::3.2.1.35",
117
+ "av::ec::EC::3.5.1.98",
118
+ "av::ec::EC::3.5.4.4",
119
+ "av::ec::EC::3.6.1.1",
120
+ "av::ec::EC::3.6.1.5",
121
+ "av::ec::EC::4.2.1.11",
122
+ "av::ec::EC::5.1.1.1",
123
+ "av::ec::EC::5.1.1.8"
124
+ ],
125
+ "lost_share": 0.08605341246290801,
126
+ "singleton_units": 93293,
127
+ "units": 110111,
128
+ "units_per_split": 2202.0280324272494,
129
+ "well_formed": true
130
+ },
131
+ {
132
+ "failures": [],
133
+ "labels_carried": 6066,
134
+ "labels_lost": 0,
135
+ "largest_unit": 1,
136
+ "level": 0.9,
137
+ "locked_share": 0.0,
138
+ "lost_labels": [],
139
+ "lost_share": 0.0,
140
+ "singleton_units": 137662,
141
+ "units": 137662,
142
+ "units_per_split": 2753.0,
143
+ "well_formed": true
144
+ }
145
+ ],
146
+ "markdown": "| Level | Units | Singleton units | Largest unit | Locked share | Units per split | Labels lost | Well formed |\n|---|---:|---:|---:|---:|---:|---:|---|\n| 0.3 | 67,767 | 47,204 | 187 | 0.0% | 1,355 | 1242 of 6066 | no: labels_lost |\n| 0.5 | 89,994 | 69,078 | 93 | 0.0% | 1,800 | 886 of 6066 | no: labels_lost |\n| 0.7 | 110,111 | 93,293 | 77 | 0.0% | 2,202 | 522 of 6066 | yes |\n| 0.9 | 137,662 | 137,662 | 1 | 0.0% | 2,753 | 0 of 6066 | yes |",
147
+ "max_unit": 275,
148
+ "proteins": 168302,
149
+ "recommended": 0.5,
150
+ "settings": {
151
+ "max_locked_share": 0.1,
152
+ "max_lost_label_share": 0.1,
153
+ "max_unit_share": 0.1,
154
+ "min_label_units": 20,
155
+ "min_units_per_split": 50
156
+ },
157
+ "small_dataset": false
158
+ }
tables/assertions.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7131ae1f14799cc94589acd13f53ab0b0504f6c7ee7c9e3d77717104210c7d7d
3
+ size 1487838289
tables/clusters.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c00cbbb414fb2221c3d13a633719db2296d2e9c8a3fea46c024e675b9dbe9690
3
+ size 396804531
tables/concepts.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:079bda9b4e14c772a4a2c2262f8a6d17300974fc5eab71b4654da5f0c7521e46
3
+ size 4914720
tables/hierarchy.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ecb329a21e70cff25c054859f58c31476cddea2ca62d1bf6bba0d95a17667c40
3
+ size 1065229
tables/measurement_texts.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:feb25e0cdd53c82536152168e9fa458d14743f0403dcf64b27d72f2cb6dc49a2
3
+ size 1502638
tables/neighbours.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8876b66b8dcee5c8382cb43130e1012ba091de07c5cf1737a563d78ecb75337f
3
+ size 58179018
tables/proteins.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7473ff99625472f3e64811256cfe989ae5e0a4824f9b1576982ecc8f432ebf39
3
+ size 535426196
tables/split.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:86ebdc7253c748756a82c1f620a57a0ba0ce8e23ca57b4afbf98de6117562ed7
3
+ size 47909341
tables/tokens.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:97223dd80ca2be25adce67446eb11c904343fae7faf866107919f640c3694191
3
+ size 1014134
tables/vocabulary.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bd288444b33d54ff26851da3b85d662d630e8ea9e3ccd99c9db0c614a6057d5e
3
+ size 8002692
test.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:74f0996a82c3b96e193b3718ae576dbcab5f9e83e39b5b1dc6abd33d04eb10a2
3
+ size 13653827
train.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0743d9aa42d6d9cf3a7c86abc586b1703e0e14f4e334992e60fa8f700e0de5fa
3
+ size 786412065
validation.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:be0e75e6ae0a5a79cda1010c9c271eee0eb99ae9b557371e2c5d212d5e7ee082
3
+ size 8102308