Datasets:

tonynzh2
/

afdb

license: other
license_name: alphafold3-output-terms-of-use
license_link: https://github.com/google-deepmind/alphafold3/blob/main/OUTPUT_TERMS_OF_USE.md
task_categories:
  - other
tags:
  - biology
  - protein-structure
  - alphafold3
  - ancestral-sequence-reconstruction
  - gene-ontology
size_categories:
  - 100K<n<1M

afdb — AlphaFold 3 predicted structures for the GO + ASR sequence catalog

Predicted monomer structures for 292,804 unique protein sequences: 38,934 from a Gene Ontology molecular-function coupling set and 254,182 audited ancestral sequence reconstructions (312 sequences occur in both). The predictions serve as teacher labels for 1,421,494 training pairs; one structure is reused by every pair referencing that exact sequence.

Sequence IDs are seq_<sha256(sequence)[:20]>, so a prediction can always be matched back to its exact sequence.

Protocol

Stock open-source AlphaFold 3 with default settings — no protocol deviations:

Setting	Value
Code	google-deepmind/alphafold3 v3.0.4 (`c0f97ed`)
Genetic search	full standard pipeline: UniRef90, MGnify, small BFD, UniProt
Templates	searched (PDB seqres + mmCIF, `max_template_date` default 2021-09-30)
Diffusion samples	5 (default)
Recycles	10 (default)
Model seed	20260916
Databases	AlphaFold 3 paper versions

Both MSAs AlphaFold 3 builds are retained: the unpaired MSA and the paired (UniProt) MSA. The paired MSA is kept even though every input is a single chain, because AlphaFold 3 featurises it as msa_all_seq regardless of chain count and its presence also changes the unpaired crop size — omitting it would not be the standard protocol.

Layout

Per shard (590 shards, ≤500 sequences each, grouped by sequence length):

predictions/<band>/<part>.tar    predicted structures, confidences, *_data.json
msa/<band>/<part>.tar            unpaired and paired A3M files

Files are bundled per shard rather than stored loose: 292,804 predictions would otherwise be well over a million objects.

Terms of use

These structures are AlphaFold 3 Output and are provided under, and subject to, the AlphaFold 3 Output Terms of Use. See LEGALLY_BINDING_TERMS_OF_USE.txt. Each per-shard archive also carries AlphaFold 3's own TERMS_OF_USE.md.

Released openly for non-commercial research.

Citation

Abramson, J. et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature 630, 493–500 (2024). https://doi.org/10.1038/s41586-024-07487-w