Unified Protein-Small Molecule Graph Neural Networks for Binding Site Prediction

Abstract

Predicting small molecule binding sites on proteins remains a key challenge in structure-based drug discovery. While AlphaFold3 has transformed protein structure prediction, accurate identification of functional sites such as ligand binding pockets remains a distinct and unresolved problem. Graph neural networks have emerged as promising tools for this task, but most current approaches focus on local structural features and are trained on relatively small datasets, limiting their ability to model long-range protein-ligand interactions. Here, we develop YuelPocket, a new graph neural network that addresses these limitations through an innovative global graph design featuring a virtual joint node connecting all protein residues and small molecule atoms to capture long-range interactions while maintaining linear computational complexity. Trained on the comprehensive MOAD dataset containing over 38,000 protein-small molecule complexes, YuelPocket achieves exceptional performance with AUC-ROC values of 0.85 and 0.89 on COACH420 and Holo4k datasets, respectively.