PSDTA: An Approach to Drug-Target Binding Affinity Prediction by Integrating Physicochemical and Structural Information to Reduce Feature Redundancy.
Accurate prediction of drug-target binding affinity (DTA) is critical for repurposing. Although deep learning has been widely applied in this field, existing methods still face challenges, including inadequate integration of physicochemical properties and structural information, which leads to high feature redundancy and limited generalization performance. To address these challenges, we propose a DTA prediction method, PSDTA (where P represents physicochemical properties and S represents structural information), which integrates physicochemical properties into the initial feature representations. Unlike conventional approaches, PSDTA explicitly incorporates structural information on amino acids, thereby avoiding the risk of information leakage caused by directly using coordinates as features and enhancing the model's generalization capability. Furthermore, two complementary channels are designed to identify binding-relevant residues at both residue and group levels, thereby reducing feature redundancy. To evaluate its effectiveness, we conducted experiments on three benchmark data sets: PDBBind v2016, PDBBind v2020, and Davis. By comparing with several state-of-the-art algorithms, PSDTA achieves the best results in terms of performance. Interpretability analyses indicate that the two channels identify highly consistent and complementary binding-related regions.