PPIformer

Weights and preprocessed datasets for PPIformer. See the GitHub page for details.

File description

weights.zip stores the weights of the trained models in the form of PyTorch Lightning checkpoints.
- The masked_modeling.ckpt stores the PPIformer model pre-trained on the PPIRef50K dataset via structural masked modeling.
- The ddg_regression folders stores 3 fine-tuned models resulting from the 3-fold cross-validation on the SKEMPI v2.0 ddG dataset.

.pyg_dataset_cache.zip stores the preprocessed PPIRef50K and SKEMPI v2.0 datasets as PyTorch Geometric tensors. If you download these files, you do not have to reconstruct the dataset from .pdb files from scratch. Please note that the subsets with the fold value (the file names are in the format ppi_inmemery_dataset_<split>_<fold>.pt) of 8 are 8 random samples useful for debugging.