Closed DimaMolod closed 2 weeks ago
there is also no 'ptm' and 'iptm' scores, the way it was: dict_keys(['distogram', 'experimentally_resolved', 'masked_msa', 'predicted_aligned_error', 'predicted_lddt', 'structure_module', 'plddt', 'aligned_confidence_probs', 'max_predicted_aligned_error', 'ptm', 'iptm', 'ranking_confidence'])
there is also no 'ptm' and 'iptm' scores, the way it was: dict_keys(['distogram', 'experimentally_resolved', 'masked_msa', 'predicted_aligned_error', 'predicted_lddt', 'structure_module', 'plddt', 'aligned_confidence_probs', 'max_predicted_aligned_error', 'ptm', 'iptm', 'ranking_confidence'])
I see. It seems 'seqs' was never here anymore. I think create_notebook.py still has to be updated.
Actually, I rerun it again and 'seqs' is among the keys ['distogram', 'experimentally_resolved', 'masked_msa', 'predicted_aligned_error', 'predicted_lddt', 'structure_module', 'plddt', 'seqs', 'unrelaxed_protein', 'aligned_confidence_probs', 'max_predicted_aligned_error', 'ranking_confidence']
Actually, I rerun it again and 'seqs' is among the keys ['distogram', 'experimentally_resolved', 'masked_msa', 'predicted_aligned_error', 'predicted_lddt', 'structure_module', 'plddt', 'seqs', 'unrelaxed_protein', 'aligned_confidence_probs', 'max_predicted_aligned_error', 'ranking_confidence']
hmm I just checked a random pkl from the prediction made today using this branch and there is no 'seqs', must be some recent fix?.. ['distogram', 'experimentally_resolved', 'masked_msa', 'num_recycles', 'predicted_aligned_error', 'predicted_lddt', 'structure_module', 'plddt']
but I didn't update alphafold submodule, so if you introduced some changes there this could be the reason
Interesting. Perhaps it's better to update create_notebook.py so that it parses sequences from PDB files directly anyway
Yes, you can do this, but we need to have ptm and iptm (as well as the rest of the missing keys) in the result.pkl anyways
ah I found where it came from. It's not from the recalculations of metrics but rather here: https://github.com/KosinskiLab/AlphaPulldown/blob/b791e1b65cff7583cd28c9fcb08e188a46db077e/alphapulldown/folding_backend/alphafold_backend.py#L397-L403 the pickle was dumped before the dictionary got updates on the missing keys.
Plus the calculation of iptm iptm+ptm etc were moved to post processing after your restructuring
Good catch!
I think after this line: https://github.com/KosinskiLab/AlphaPulldown/blob/main/alphapulldown/folding_backend/alphafold_backend.py#L542 there is no 'ranking_confidence' in the result.pkl file, after recalculate_confidence