Using SPEACH-AF with AF2 instead of ColabFold

[JasonWang2002]

Hello,

The .ipynb script works correctly for .a3m files generated from ColabFold; however, when I try to use the generated files in AlphaFold2 version 2.3, I get the following error message:

Traceback (most recent call last):
File "/path/to/alphafold232/alphafold/run_alphafold.py", line 468, in
app.run(main)
File "/path/to/alphafold232/miniconda3/envs/af232/lib/python3.8/site-packages/absl/app.py", line 312, in run
_run_main(main, args)
File "/path/to/alphafold232/miniconda3/envs/af232/lib/python3.8/site-packages/absl/app.py", line 258, in _run_main
sys.exit(main(argv))
File "/path/to/alphafold232/alphafold/run_alphafold.py", line 443, in main
predict_structure(
File "/path/to/alphafold232/alphafold/run_alphafold.py", line 196, in predict_structure
feature_dict = data_pipeline.process(
File "/path/to/alphafold232/alphafold/alphafold/data/pipeline.py", line 232, in process
msa_features = make_msa_features((uniref90_msa, bfd_msa, mgnify_msa))
File "/path/to/alphafold232/alphafold/alphafold/data/pipeline.py", line 79, in make_msa_features
features['deletion_matrix_int'] = np.array(deletion_matrix, dtype=np.int32)
ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (12654,) + inhomogeneous part.

This error does not arise when using the _00.a3m file generated that is a copy of the input .a3m, but does for the other output .a3m files which are edited. Could you help us understand the cause of this issue?
 
Thanks,
Jason

[iamysk]

I've got the same issue. Only protein_00.a3 (original MSA) seems to work. And I see this error both with AF2 and colabfold.

ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (5064,) + inhomogeneous part

This is for a revision, so it'd be great if a patch or solution could be provided.

Thanks