Encoding Dynamic Information from Normal Mode Analysis in Graph Representation Learning for Protein Function Prediction

The relationship between protein structure and molecular function has fundamental importance for biological science and pharmaceutical application but remains challenging to probe in by experimental approaches. Recent advances in protein function prediction exploit graph-based deep learning approaches to correlate the structural and topological features of proteins with their molecular functions. However, proteins <i>in vivo</i> are not static but dynamic molecules that alter conformation for functional purposes. We recently applied normal mode analysis to native protein conformations and augment protein graphs by connecting edges between dynamically correlated residue pairs. In the multilabel function classification task, our method secures a remarkable performance gain based on the dynamics-informed representation. The proposed graph neural network, ProDAR, increases the interpretability and generalizability of residue-level annotations and robustly reflects structural nuance in proteins. We elucidate the importance of dynamic information in graph representation by comparing gradient-weighted class activation maps with and without dynamic information as inputs. Our model learns the dynamic fingerprints of proteins and pinpoints the residues of functional impacts that hold potential interests for protein and drug design.