Hydrogen-Bond Graph Neural Network Improves Protein Representation Learning

A graph neural network for protein modeling constructs edges from energetically filtered hydrogen bonds and augments node features with secondary structure assignments, capturing both local motifs and long-range couplings beyond simple sequence or proximity graphs.