jozhang97 / mutateeverything Goto Github PK

Hi authors, thanks for the great work!

When I try to reproduce the test demo results of AlphaFold-based model with the provided command, I meet mismatch between the results on my side and the results in this repo. Cound you kindly suggest the possible reason? I follow your instruction to set up the environment on a A100 server with CUDA-11.4. You can download my results to see the clear difference, and here are some intuitive comparisons:

My single mutation predictions:

Single mutation predictions in this repo:

Is the positive ddg value in the dataset a good mutation or a bad mutation?

Missing openfold/data and openfold/resources?

When I was observing the dataset, I found that cdna test dataset is all double mutation. Is there a test dataset for single mutation?

Hello,
I am having a trouble when loading the MutateEverything af model.
it seems like there are some mismatches between the given weights and the definition of the model in the code.
Traceback (most recent call last):
File "/root/capsule/code/MutateEverything-master/test.py", line 179, in
main(args)
File "/root/capsule/code/MutateEverything-master/test.py", line 102, in main
misc.load_model(args, model, None, None)
File "/root/capsule/code/MutateEverything-master/misc.py", line 385, in load_model
model_without_ddp.load_state_dict(checkpoint['model'])
File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1604, in load_state_dict
raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for MutateEverything:
Unexpected key(s) in state_dict: "backbone.bb_adapter1.linear1.weight", "backbone.bb_adapter1.linear1.bias", "backbone.bb_adapter1.linear2.weight", "backbone.bb_adapter1.linear2.bias", "backbone.bb_adapter1.norm.weight", "backbone.bb_adapter1.norm.bias".

I would Highly appreciate your guidance and help! Thx!

Thank you for sharing your code and data :)
However, I encountered an issue while examining the cDNA data. I noticed that several PDB entries, which provide corresponding MSAs and FASTA files, are not listed in cdna_train.csv or cdna2_test.csv.

Could it be possible that the mutation data records for these PDB entries were omitted during the compilation process? (Here are the PDB codes):

6scw, 5ubs, 5uce, 3cqt