Initiated integration of ColabFold-AF3 pipeline. Draft state of AF3 json input generation.

[rachelse]

Major

batch.py

• Added argument group to handle specific arguments for AF3.
  • --af3-json to specify AF3 json generation.
  • --fasta-type to handle non-protein molecules in fasta file.
    The header should be > entityID | entityType | numberOfEntity
    e.g. FASTA file with 1 homodimer and 3 ligand

    > Prot1|protein|2
    PROTEINSEQUENCE
    > Lig1|ccd|3
    ATP
    

• Added functions
  • generate_af3_input processes fasta and a3m files and produce json files if specified --af3-json. (This will not proceed with structure inference.)
  • get_queries_af3 handles fasta files containing complex with non-protein molecules if --fasta-type is 1.
  • parse_fasta_af3 is called by get_qureries_af3 to parse single fasta file .

utils.py

• Added class AF3Utils to fill out elements in json file.

Subsidiary

• msa2json.py

  This code was adopted from Yoshi's code. It generates json inputs from given a3m file / directgory contains a3m files.

• af3json.py

  Adaptation of msa2json.py to generate default AF3 json format. This json will make AF3 to produce MSAs using their own pipeline.

Progress

• Convert fasta/msa to json
  • colabfold_batch
  • colabfold_search
• Support FASTA file with multiple molecule types
  • single fasta file per complex
  • directory with multiple fasta files
• Support template
• Support unpairedMsaPath and pairedMsaPath
• Test pipeline to validate json file
  (each molecule types, msa-pairing, template, custom ccd, RNA MSA, modifications, etc.)
• Validation
  • molecule types
    • single protein, protein complex, protein+ligand
    • protein+RNA, protein+DNA, RNA, DNA, ligand
  • msa-pairing
    • UnpairedMSA/PairedMSA
    • dependency on --pair-mode
    • specify with path to MSA file