Geometric Features and GAT Neural Network for Protein Surface Classification

Proteins are dynamic macromolecules with evolving 3D shapes, making classification challenging. We introduce a novel method that estimates local principal curvatures at each mesh vertex to capture key geometric features of protein surfaces. Features are represented as a graph and classified using double-headed graph attention networks (GATs). Our approach emphasizes the importance of surface characteristics in biological function, offering a systematic framework for protein surface analysis and classification.