PLANET: A Multi-objective Graph Neural Network Model for Protein–Ligand Binding Affinity Prediction

• DOI: 10.1021/acs.jcim.3c00253
• Publication Date: 2023-06-15T20:36:06Z
• Authors: Xiangying Zhang; Haotian Gao; Haojie Wang; Zhihang Chen; Zhe Zhang; Xinchong Chen; Yan Li; Yifei Qi; Renxiao Wang
• Keywords: Docking (animal); Benchmark (surveying); Protein ligand

Predicting protein–ligand binding affinity is a central issue in drug design. Various deep learning models have been published recent years, where many of them rely on 3D complex structures as input and tend to focus the single task reproducing affinity. In this study, we developed graph neural network model called PLANET (Protein–Ligand Affinity prediction NETwork). This takes graph-represented structure pocket target protein 2D chemical ligand molecule input. It was trained through multi-objective process with three related tasks, including deriving affinity, contact map, distance matrix. Besides complexes known data retrieved from PDBbind database, large number non-binder decoys were also added training for final PLANET. When tested CASF-2016 benchmark, exhibited scoring power comparable best result yielded by other well reasonable ranking docking power. virtual screening trials conducted DUD-E PLANET's performance notably better than several machine models. As LIT-PCBA achieved accuracy conventional program Glide, but it only spent less 1% Glide's computation time finish same job because did not need exhaustive conformational sampling. Considering decent efficiency prediction, may become useful tool conducting large-scale screening.