Co-guided Dual-channel Graph Neural Networks for the prediction of compound–protein interaction

Abstract

Compound–Protein Interaction (CPI) serves as essential indicators for efficiently screening potential candidate drugs. Previous studies have typically focused on modeling CPIs either from intramolecular or intermolecular interactions, disregarding the diversity of interactions and the fine dependencies between these two types of interactions, thereby limiting the accuracy of CPI predictions. We argue that properly considering both intramolecular and intermolecular interactions allows for a more comprehensive understanding of the interactions between compounds and proteins. To this end, we propose a novel approach called Co-guided Dual-channel Graph Neural Network (CDGN) for CPI predictions. CDGN simultaneously captures various CPI information from intramolecular and intermolecular interactions using a dual-channel aggregating mechanism. Furthermore, to model the complicated relationships between the two interactions, we design a co-guided learning scheme to model the CPIs between intramolecular and intermolecular interactions, enhancing the learning of each other. Finally, we predict CPIs based on the rich interaction information from dual channels. Exhaustive experimental studies on two benchmarks verify the superiority of CDGN in CPI predictions. In particular, CDGN achieves outstanding performance with RMSE evaluation metrics of 1.263 and 1.626 on the publicly available PDBbind and CSAQ-HiQ datasets, respectively.