Database usage for AlphaFold3 runs

[dmora127]

Hi folks,

I skimmed through the repo and it looks like the create_individual_features.py script is expecting the following databases:

AF3_DATABASES = {
"uniref90": "uniref90_2022_05.fa",
"uniref30": "uniref30/UniRef30_2023_02",
"mgnify": "mgy_clusters_2022_05.fa",
"bfd": "bfd/bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt",
"small_bfd": "bfd-first_non_consensus_sequences.fasta",
"pdb_seqres": "pdb_seqres_2022_09_28.fasta",
"template_mmcif_dir": "mmcif_files",
"obsolete_pdbs": "obsolete.dat",
"pdb70": "pdb70/pdb70",
"uniprot": "uniprot_all_2021_04.fa",
"ntrna": "nt_rna_2023_02_23_clust_seq_id_90_cov_80_rep_seq.fasta",
"rfam": "rfam_14_9_clust_seq_id_90_cov_80_rep_seq.fasta",
"rna_central": "rnacentral_active_seq_id_90_cov_80_linclust.fasta",
}

It doesn't seem to be used anywhere else in the repo. AlphaFold3 typically does not run using uniref30/UniRef30_2023_02, bfd/bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt, obsolete.dat, or pdb70/pdb70. Is this a typo or did I accidentally miss anything? 😅

We're trying to use the AlphaFold3 run config due to its reduced data storage footprint on our system, and it would be helpful to know if those are actually being used or if they were copied over by accident.

[DimaMolod]

Hi,

I checked the AF3 code path in create_individual_features.py, and I do not think you missed anything.

In the current implementation, the AF3 pipeline only passes the following databases into AF3DataPipelineConfig: small_bfd, mgnify, uniprot, uniref90, ntrna, rfam, rna_central, template_mmcif_dir, and pdb_seqres.

I do not see uniref30, bfd, pdb70, or obsolete_pdb being used anywhere in the AF3 pipeline.

So seems like you are right, and those entries in AF3_DATABASES are just leftovers from the AF2-style database map rather than actual AF3 requirements. I think they can be safely removed.

Thanks for spotting this!