ESM3-PPISites: Decoding the Grammar of Protein-Protein Interaction Interfaces with Multimodal Representations

The data in this directory is part of the published paper Decoding the Grammar of Protein–Protein Interaction Interfaces with Multimodal Representations. It contains the dataset splits required for model training, validation, and evaluation results. Uniprot ids are used as identifiers within the datasets. esm3_data/: Root directory for ESM3 protein embeddings and metadata. big_model_data/: PyTorch tensor files (.pt) from the large model (98B parameters), split into train/val/test sets for BioLiP-3693 and PDBbind-1409, plus a ZK448 test set. small_model_data/: PyTorch tensor files (.pt) from the small model (1.4B parameters), using the exact same datasets and split structure as the big model. csv_data/: Tabular files (.csv) matching the names and splits of the tensor data to provide corresponding labels.