The manuscript scores P(catalytic metal) with Metal3D-Cat. That is just another --models entry. The same wrapper runs Metal3D (any metal site) or Metal3D-Cat (catalytic metal site).
metal3d_model scans through the current PDB and predicts based on the local chemical enviornments the probability and position of a metal site. It writes pmetal (higher is better).
Same yaml block. Only model_path changes.
| weights | use when |
|---|---|
metal3d_cat.pth |
Metal3D-Cat for predicting a catalytic metal (the paper) |
metal3d_clean.pth |
Metal3D-Clean for predicting a generic metal site |
models:
metal3d:
model_path: ./models/metal3d/weights/metal3d_cat.pth
max_metal_p: 0.2
threshold: 5The CLI name is metal3d_model. The config section is metal3d.
python run.py -c configs/your_run.yml \
--models seq_model metal3d_modelAdd pocket shape, physics, pKa, or anything else on the same line. A demo that combines both Metal3D-Cat and pocket shape is available in example_substrate.md.
| file | role |
|---|---|
models/metal3d_model.py |
the dEVA wrapper |
models/metal3d/weights/metal3d_cat.pth |
Metal3D-Cat |
models/metal3d/weights/metal3d_clean.pth |
Metal3D-Clean |
models/metal3d/weights/metal_0.5A_v3_d0.2_16Abox.pth |
Metal3D OG |