kuhlman-lab / alphafold Goto Github PK

Hello.
Thanks for this awesome work.
I am trying to model the complexes of a few hundred peptides with a multichain target.
I couldn't figure out how to use multichain custom MSAs. Could you please point me towards some documentation/examples regarding this?
To bypass this, I connected my multi (two) chain target with a polyglycine linker.
My af2_flags file looks like this

--input_dir input
--output_dir AFmult/predictions
--params_dir /home/amin/softwares/Protein-Design/params-2022-12-06/
--compress_output
--use_ptm
--msa_mode single_sequence
--custom_msa_path /home/amin/Work/Project_42/msas_complex
--max_recycle 4
--save_timing
--num_models 5
--num_seeds 5

When I start the calculation, it uses ~100 GB of RAM and takes a lot of time, approximately 3 hours, before it starts generating models. Am I doing something wrong here? Are there some example of launching such jobs on multi-GPU workstations? Or is this delay and memory usage expected for a 750 residue system where the MSA of 735 residues has been precomputed?
If you prefer, I can open the two issues separately.
I would be really grateful for any suggestions.
Best,
Amin.