matthiasblum/uniparc-interpro

UniParc 2026_03 with InterPro 110.0 annotations

All protein sequences of UniParc (UniProt release 2026_03), with the matches found by InterProScan 6 for InterPro release 110.0.

• Sequences: 1,200,813,736
• Matches: 19,096,091,782

Matches come from two sources:

• InterPro: InterProScan runs each member database's own models (HMMs, profiles, patterns, etc.) and reports the signatures that match.
• InterPro-N: a deep-learning model developed by Google DeepMind that predicts member database signatures directly from the sequence, without the databases' own models. It can find matches that the classical methods miss. InterPro-N matches are reported separately (source = 'InterPro-N'). They are not integrated into InterPro entries and have no E-value; each location has a score between 0 and 1 (the model's confidence).

Configs

sequences

One row per UniParc sequence, sorted by UPI.

Column	Type	Description
upi	string	UniParc identifier
md5	string	MD5 checksum of the sequence (uppercase hex)
crc64	string	CRC64 checksum of the sequence
length	int32	Sequence length
created	date32	Date the sequence was added to UniParc
sequence	string	Amino acid sequence

annotations

One row per match (a signature matching a sequence), sorted by MD5. Sequences without any match have no rows.

Column	Type	Description
upi	string	UniParc identifier
md5	string	MD5 checksum of the sequence
source	string	Method that produced the match: InterPro (InterProScan, with the member database's own models) or InterPro-N (deep-learning predictions)
database	string	Member database of the signature (e.g. Pfam, PANTHER)
database_version	string	Member database version; it can differ between InterPro and InterPro-N (see Member databases)
signature_accession	string	Signature accession (e.g. PF00069)
signature_name	string	Signature short name
signature_description	string	Signature description
signature_type	string	Signature type (Family, Domain, Repeat, ...)
entry_accession	string	InterPro entry integrating the signature, if any
entry_name	string	InterPro entry short name
entry_description	string	InterPro entry name
entry_type	string	InterPro entry type
score	float64	Match score (database-dependent)
evalue	float64	Match E-value (database-dependent)
locations	list	Match locations: start, end, score, evalue, pvalue, sequence_feature, and fragments (list of start, end, dc_status)

Positions are 1-based and inclusive.

Member databases

Versions of the member databases, as reported in database_version. InterPro-N (model version 105.0) predicts signatures from older releases of some databases. A signature found by both sources appears twice for the same sequence, once per source, possibly with different versions. Databases marked — are only reported by InterPro.

Database	InterPro	InterPro-N
AntiFam	8.0	-
CATH-FunFam	4.3.0	-
CATH-Gene3D	4.3.0	4.3.0
CDD	3.21	3.21
COILS	2.2.1	-
HAMAP	2026_01	2025_01
MobiDB-lite	4.0	-
NCBIFAM	20.0	17.0
PANTHER	19.0	19.0
Pfam	38.2	37.3
Phobius	1.01	-
PIRSF	3.10	3.10
PIRSR	2025_05	-
PRINTS	42.0	42.0
PROSITE patterns	2026_01	2025_01
PROSITE profiles	2026_01	2025_01
SFLD	4	4
SignalP	6.0i	-
SMART	9.0	9.0
SUPERFAMILY	1.75	1.75
TMbed	1.0.2	-

Usage

from datasets import load_dataset

sequences = load_dataset("matthiasblum/uniparc-interpro", "sequences", split="train", streaming=True)
annotations = load_dataset("matthiasblum/uniparc-interpro", "annotations", split="train", streaming=True)

Annotation files are sorted by MD5 and split by its first byte (annotations/part-XX-*.parquet), so the matches of one sequence can be read remotely with DuckDB by querying only its shard:

SELECT source, database, database_version, signature_accession, entry_accession, locations
FROM 'hf://datasets/matthiasblum/uniparc-interpro/annotations/part-EF-*.parquet'
WHERE md5 = 'EF8A186543FE2E2243B5F2C571E8CE69';

Queries over all files (e.g. finding a UPI in sequences/, or joining both configs) read every file's footer and are slow over HTTP. Download the dataset first:

hf download matthiasblum/uniparc-interpro --repo-type dataset --local-dir uniparc-interpro

Then, from uniparc-interpro/:

SELECT s.upi, s.length, a.source, a.database_version, a.signature_accession, a.entry_accession
FROM 'sequences/*.parquet' s
JOIN 'annotations/*.parquet' a USING (md5)
WHERE s.upi = 'UPI0000000001' AND a.database = 'Pfam';

References

If you use this dataset, please cite the following resources:

• UniProt Consortium. UniProt: the Universal Protein Knowledgebase in 2025. Nucleic Acids Res. 2026. DOI: 10.1093/nar/gkae1010
• Blum, et al. InterPro: the protein sequence classification resource in 2025. Nucleic Acids Res. 2025. DOI: 10.1093/nar/gkae1082
• Blum, et al. InterProScan 6: a modern large-scale protein function annotation pipeline.Bioinform. Adv. 2026. DOI: 10.1093/bioadv/vbag141