MaSIF

MaSIF extracts interaction fingerprints from protein molecular surfaces using geometric deep learning to predict biomolecular interactions from surface chemical and geometric features.

Key Features:

• Geometric Deep Learning: Employs geometric deep learning algorithms to learn patterns on protein molecular surfaces that underlie biomolecular interactions.
• High-Level Surface Representation: Abstracts protein structures to molecular-surface representations that emphasize interaction-relevant chemical and geometric features.
• Pattern Recognition Across Evolutionary Histories: Identifies common interaction fingerprints among proteins with similar interaction modes regardless of evolutionary relationships.
• Learning from Large-Scale Datasets: Trains models on extensive datasets to recognize interaction patterns not apparent from manual inspection of structures.

Scientific Applications:

• Protein Pocket-Ligand Prediction: Identifies potential ligand-binding pockets on protein surfaces to inform drug discovery and ligand docking hypotheses.
• Protein-Protein Interaction Site Prediction: Predicts surface regions likely to mediate protein-protein interactions to aid study of cellular mechanisms and pathways.
• Ultrafast Scanning of Protein Surfaces: Rapidly scans protein surfaces to predict formation of protein-protein complexes and to enable large-scale surface comparisons.

Methodology:

Applies geometric deep learning to extract and analyze interaction fingerprints from protein molecular surfaces and trains models on large-scale structural datasets.