google-deepmind / google-deepmind/alphafold
Function about MAP_HHBLITS_AATYPE_TO_OUR_AATYPE
- Dominant language
- Python
- Stars
- 14.9k
- Forks
- 2.9k
- PR merge metrics
- No merged PRs in 30d
Description
In monomer inference, there are two function, _**alphafold.model.tf.data_transforms.correct_msa_restypes**_ and **alphafold.model.tf.data_transforms.fix_template_aatype**, to correct msa restype and template_aatype to have the same order as residue_constants, respectively.
In multimer inference, before the same operation as monomer inference in multimer inference, there is **alphafold.data.feature_processing._correct_msa_restypes** to correct msa restype to have the same order as residue_constants.
In my understanding, you did twice correct_msa_retypes in multimer inference. One is in the msa pairing and another is in process_features. Is it right? Is it specific for multimer training and inference?
Actually, I have tested that difference times for correcting msa restypes produce different outputs.
Contributor guide
Assessment
This issue has not been assessed yet.