Guidance on Predicting Protein-Peptide Complexes Using AlphaFold3

[ACIITB23]

I am currently facing an issue with predicting protein-peptide complexes. I have successfully installed AlphaFold3 locally and received the model parameters. While I am able to predict individual protein structures using the appropriate commands, I am encountering difficulties when trying to predict complexes. I have tried modifying the JSON files, but I was unable to run the predictions successfully.

Could you kindly provide the steps and commands required for predicting protein-peptide complexes? Additionally, I would appreciate guidance on how to handle predictions for more than 100 sequences.

Thank you very much for your assistance!

[zhanglzu]

This is a JSON file for predicting protein complexes that I used. Maybe it was not entirely correct. For reference only.

{
    "name": "pos_v4",
    "modelSeeds": [
        10,
        42
    ],
    "sequences": [
        {
            "protein": {
                "id": "E",
                "sequence": "MSSTHSNNVGHPQSSPQGPLTEQQRAQQQYQIFENSLPKVSQSVYQMLLNEMVPLAMGIERQISGDVISSDSNVTSENGNINNMIKRLKIEEHHTVDIIRSHNLIHELYKADEEEKEKVLARLRNIGFQIGLKLSELLIFSNNPNLKFKEMDLLLIMKFICRDVWKQIFGKQIDNLKTNHRGTFYLLDYDYRPIQSFSLEEDAKNEELKMIEPFLEIPVGIIRGVLSSLGYSSEEVICLASFIDRPTDRPKTAFPKGVSFHVQVTMPQ"
            }
        },
        {
            "protein": {
                "id": "P",
                "sequence": "MVSTTQSRSLKAMGEEIWKNKTEKINTELFTLTYGSIVAQLCQDYERDFNKVNDHLYSMGYNIGCRLIEDFLARTALPRCENLVKTSEVLSKCAFKIFLNITPNITNWSHNKDTFSLILDENPLADFVELPMDAMKSLWYSNILCGVLKGSLEMVQLDCDVWFVSDILRGDSQTEIKVKLNRILKDEIPIGED"
            }
        }
    ],
    "dialect": "alphafold3",
    "version": 1
}

[Augustin-Zidek]

Agreed with the comment above -- specify your two chains using the AlphaFold 3 JSON format and pass path of the input JSON file as a flag to AlphaFold. Refer also to https://github.com/google-deepmind/alphafold3/blob/main/docs/input.md for more details.

Additionally, I would appreciate guidance on how to handle predictions for more than 100 sequences.

Specify the 100+ sequences in the sequences field. In case some have multiple copies, you can specify them just once and use the multi-id functionality: "id": ["A", "B", "C", ...].