MGAOD-DTI: multihead graph attention and omni-dimensional dynamic convolution-based method for drug-target interaction prediction

Identifying the interactions between drugs and targets plays a pivotal role in drug discovery and development, as it helps uncover binding relationships that can significantly reduce research costs and accelerate development timelines. As an essential step in this process, Drug-Target Interaction (DTI) prediction has garnered considerable attention, with recent advances in deep learning-based methods showing notable improvements in integrating drug molecular structures and target sequence features. Despite these advancements, existing approaches often struggle to capture the intricate characteristics of drug molecules and the key binding sites of targets with sufficient accuracy. To address these challenges, this study proposes a novel model, the multi-head graph attention and omni-dimensional dynamic convolution-based method for drug-target interaction prediction (MGAOD-DTI), which incorporates multi-head graph attention and omni-dimensional dynamic convolution. By leveraging a dynamic mechanism, the model combines the two-dimensional (2D) substructural features and three-dimensional (3D) spatial features of drugs, along with target sequence features, to enhance its ability to represent critical binding sites and improve prediction accuracy. Experimental evaluations reveal that MGAOD-DTI achieves superior performance compared to state-of-the-art methods across multiple metrics on the Human, C.elegans, and KIBA datasets. These results underscore the robustness and generalizability of MGAOD-DTI, offering a powerful and scalable solution for advancing DTI prediction in drug development.