gitter-lab / metl-pretrained Goto Github PK

The Windows tests are failing after the NumPy 2.0 release

The saved METL models are current in temporary storage on Box. We need to select a permanent repository and move them there. Options include hosting them on Zenodo, using Hugging Face, or something else.

The code example use the model bcEoygY3. We can replace it with one of the current models YoQkzoLD or PEkeRuxb and then uncomment the respective test.

We should consider what error handling is needed. What should happen if an the provided PDB file is inconsistent with the one a METL-Local model was trained on? What should happen if the wild type amino acid in a variant to be scored does not match the actual wild type amino acid at that position?

test3.py requres 1gfl_cm.pdb, which is not currently tracked in the repo. After we add it, we can uncomment the respective test.

We would like to publish this package on PyPI so that pretrained models can be loaded without cloning this repo. Before doing that, we need to think through how it interacts with the code in the metl repository and what the package names should be.