# interpro_repeat

Protein-level multi-label dataset sourced from InterPro entry metadata and
UniProtKB protein-to-InterPro cross-references.

## Intended use

Protein repeat annotation prediction from sequence. This evaluates
recovery of curated InterPro classifications, not an experimental assay result.

## Source and labels

- InterPro metadata: `https://ftp.ebi.ac.uk/pub/databases/interpro/current_release/entry.list`
- UniProt REST API: `https://rest.uniprot.org/uniprotkb/stream`
- Card generated (UTC): `2026-10-07`
- Organism Taxonomy ID: `9606` (`all` = all organisms).
- Review status: `reviewed` (`reviewed` = Swiss-Prot, `unreviewed` = TrEMBL, `all` = both).
- UniProt query: `(organism_id:9606) AND (reviewed:true)`
- UniProt TSV fields: `accession,sequence,xref_interpro,organism_id,reviewed`
- InterPro entry type: `Repeat`
- Consider entries observed in at least 5 proteins.
- After splitting, retain entries with at least 20
    positives in training, 3 in validation,
    and 3 in test. Because validation and test
    support select the entries, scores are conditional on that label selection.
- `targets` is a multi-hot vector ordered by InterPro accession. Accession,
  display name, and type are recorded in `label_vocabulary.json`.
- Proteins without a retained entry are dropped, both before splitting and
    again after split-support filtering, so no row has all-zero targets.
- `id` is the UniProtKB accession of the first entry carrying each unique sequence.

## Splits

Whole MMseqs2 clusters are assigned to splits targeting
`{'train': 0.8, 'validation': 0.1, 'test': 0.1}`. Seed: `1957723`. Clustering settings:

- MMseqs workflow: `easy-cluster`
- Minimum sequence identity (`--min-seq-id`): `0.3`
- Minimum coverage (`-c`): `0.8`
- Coverage mode (`--cov-mode`): `1`
- Threads: `1`
- Other MMseqs2 options use MMseqs2 defaults.

When `create_split_subsets` is enabled, pooled random, stratified (by each
protein's globally rarest positive entry), and hold-cluster-out subsets are
also included. Hold-cluster-out subsets require enough MMseqs clusters to
populate all three roles.
The optional maximum sequence length is `None` and is
applied before vocabulary construction and MMseqs2 clustering.

## Split sizes

- `test`: 116 rows
- `test_cluster`: 116 rows
- `test_random`: 116 rows
- `test_stratified`: 116 rows
- `train`: 850 rows
- `train_cluster`: 850 rows
- `train_random`: 850 rows
- `train_stratified`: 850 rows
- `validation`: 118 rows
- `validation_cluster`: 118 rows
- `validation_random`: 118 rows
- `validation_stratified`: 118 rows

Vocabulary size: 10.

## Dataset statistics

[`stats.json`](stats.json) at the dataset root contains row counts by split,
columns, and SeqKit sequence-length metrics.