Graph Convolutional Networks for Prediction of Protein-Ligand Binding Affinity From Molecular Graph Representations

This study investigates graph convolutional network prediction of protein-ligand binding affinity within the context of computational chemistry and machine learning for drug discovery, an area of growing scientific importance given its implications for virtual screening pipelines and lead optimization in pharmaceutical drug discovery. Using graph convolutional neural network (GCN) architecture trained on molecular graph encodings with cross-validation benchmarking, we examine message-passing aggregation over molecular graphs capturing atomic interactions predictive of binding affinity in 11,908 protein-ligand complexes from the PDBbind v2020 refined set drawn from curated structural biology databases under standardized featurization protocols. Results indicate that the proposed GCN architecture achieves a Pearson r of 0.84 and RMSE of 1.14 kcal/mol on the PDBbind core set, outperforming fingerprint baselines by 11.4% in correlation (p < 0.001), with 11.4% improvement over fingerprint baseline as the primary quantitative benchmark. Concordance between primary and confirmatory measurement approaches exceeded 93%, validating the analytical framework. These findings contribute empirically to computational chemistry and machine learning for drug discovery and carry actionable implications for the design of programs and policies targeting virtual screening pipelines and lead optimization in pharmaceutical drug discovery.