CoA-DTI: A Drug–Target Interaction Prediction Model Based on Co-Attention

Abstract

Accurate prediction of drug–target interactions (DTIs) is a fundamental task in AI-driven drug discovery and repositioning. While traditional experimental methods are reliable, they are often time-consuming and expensive, limiting their scalability. Recent computational approaches have shown promise but still face challenges in extracting informative representations and modeling cross-modal interactions between drugs and proteins. In this work, we propose CoA-DTI, a novel deep learning framework that integrates Graph Convolutional Networks (GCNs), Convolutional Neural Networks (CNNs), and a Co-Attention mechanism for DTI prediction. Drug molecules are encoded as molecular graphs and processed through multi-layer GCNs to capture topological and atom-level features. Protein sequences are encoded via CNNs to learn local biochemical patterns. A co-attention module is then employed to model bidirectional interactions between the drug and protein representations, enabling fine-grained and context-aware feature fusion. Extensive experiments on the benchmark Davis dataset show that CoA-DTI achieves superior performance over competitive baselines in terms of accuracy and robustness. Ablation studies further validate the effectiveness of each module. This work demonstrates the potential of integrating graph-based learning and cross-modal attention in bioinformatics applications, offering a generalizable and interpretable approach for interaction prediction.