Open s-kyungyong opened 1 year ago
Hi, can you provide a result produced by AlphaFold2? We would try to reproduce and solve it.
Hi @Gy-Lu! Some of the alphafold outputs are here. I think fastfold is treating the inference as monomer despite --model_preset multimer
. In the folder that stores precomputed alignments, there is only one folder for AT1G52380 for Fastfold. For Alphafold, the same alignments are stored separately in msas/A and msas/B. Perhaps this modification made the difference?
Hi, I have tried your sequence.
FastFold does not treat the inference as monomer. However, the two sequence have exactly same tag, and the second one would cover the first one.
With changing the second tag to AT1G523802
, I got a pdb file not equal to the alphafold ones you offered. But I think it might be caused by different weight(seems you use v3 model?) and the MSA and template searching process.
The alignment thing would be solved lately. But I am not sure the model's performance is correct, one way to examine is to use pre-computed feature and the same weight to see the output.
Yes! I was also able to model the homodimers by changing the sequence name and adding the alignments accordingly.
Wouldn't it make more sense for fastfold to automatically do this given the two same tags in a fasta file and --model_preset multimer instead of having the users change the tags?
Hi, I am not sure if there is a random in preprocessing. But it seems to be. The reason for using tag as the MSA folder name is to make pre-computed alignments easy to index. The situation of two same sequences is a little beyond our design. I think reusing computed alignments is a good thing. Following this principle, the second sequence should use the first one's alignments. Do you think this idea make sence?
Hi
While checking the multimer outputs, I realized that for homodimers, there is only a single chain, B.
python /global/scratch/users/skyungyong/Software/FastFold/inference.py --output_dir ./ --model_preset multimer --use_precomputed_alignments Alignments --enable_workflow --inplace --param_path /global/scratch/users/skyungyong/Software/FastFold/data/params/params_model_1_multimer_v3.npz --model_name model_1_multimer AT1G52380_and_AT1G52380.fasta /global/scratch/users/skyungyong/Software/alphafold-multimer-v2.2.2-080922/Database/pdb_mmcif/mmcif_files/
Alphafold v2.3.1 seems to behave as expected, generating two chains. Is this behavior expected in fastfold?