# Pfam Seed Protein Domain Classification Splits (Random and Clustered) and Pfam-N2 Predictions (Google Research)

## Facts

- Provider: Google Research (Bileschi et al.)
- License: CC0-1.0 (https://storage.googleapis.com/brain-genomics-public/research/proteins/pfam/LICENSE)
- Stored at: Google Cloud Storage, gs://brain-genomics-public/research/proteins/pfam (us)
- Access: public, readable without an account
- DOI: https://doi.org/10.1038/s41587-021-01179-w
- Category: Proteomics
- Source: https://doi.org/10.1038/s41587-021-01179-w
- Page: https://scigantic.com/archive/c0207e70-8c73-54bf-949f-f0749e476443
- Keywords: Proteomics, Google Cloud Storage

## About

Machine learning splits of the Pfam 32.0 seed alignments released with Bileschi et al., Using deep learning to annotate the protein universe (Nature Biotechnology, 2022; preprint title Can Deep Learning Classify the Protein Universe). The task is to predict which of about 18,000 Pfam families a protein domain sequence belongs to, from about 1 million training domains. Each record has the domain sequence, family_accession (the label, e.g. PF02953.15), family_id, sequence_name (UniProt accession with start and end) and the aligned_sequence from the seed multiple sequence alignment. Measured 2026-11-17: random_split/ has train/ (80 CSV shards), dev/ and test/; clustered_split/ holds a 258 MB tar.gz of a clustered train/dev/test split for harder out-of-distribution evaluation; pfam-n2/v35.0/ holds significant and insignificant domain-call tables (about 2 GB TSV each) with aligned Stockholm and unaligned FASTA archives; a models/ folder sits alongside. Load the CSVs with pandas. The folder's LICENSE and README state CC0 1.0; cite Bileschi et al. 2022.

## Load the data

```python
import gcsfs
fs = gcsfs.GCSFileSystem(token='anon')
fs.ls('brain-genomics-public/research/proteins/pfam/', detail=False)[:10]
```

Lists the files with anonymous access. Runs on any machine, with no cloud account.

## Data structure

Example file: `research/proteins/pfam/clustered_split/dev/data-00000-of-00010`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/c0207e70-8c73-54bf-949f-f0749e476443
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/c0207e70-8c73-54bf-949f-f0749e476443
- Over MCP: `get_schema_card` with id `c0207e70-8c73-54bf-949f-f0749e476443` at https://api.scigantic.com/mcp

## Related topics

- [Proteomics datasets](https://scigantic.com/topics/proteomics)
