CNNRMHSA-LGTEN: An Efficient Local Graph Transformer and Convolutional Attention Exchange Network for Drug Target Interaction Prediction

Abstract

Drug Target Interaction (DTI) prediction plays a crucial role in drug discovery by reducing experimental cost and accelerating candidate screening. Traditional computational approaches, including similarity-based and network-based methods, often fail to capture complex nonlinear relationships between molecular structures and protein sequences. Recent deep learning models improve prediction accuracy but still struggle to jointly model global molecular topology and fine-grained local chemical substructures that drive binding specificity. In this work, we propose a CNNRMHSA-LGTEN: An Efficient Local Graph Transformer and Convolutional Attention Exchange Network for Drug Target Interaction Prediction for accurate DTI prediction. The proposed framework explicitly constructs khop local concept subgraphs from drug molecular graphs to preserve functional chemical motifs, while a relative multi-head self-attention mechanism encodes both structural and attribute-level dependencies. Protein sequences are modeled using stacked one-dimensional Convolutional Neural Networks to extract hierarchical residue patterns. A gated attention-based fusion module integrates drug and protein representations, followed by a fully connected prediction head. Experimental results demonstrate that the proposed model consistently outperforms state-of-the-art baselines across multiple evaluation metrics, highlighting the effectiveness of combining local subgraph awareness with attention-driven graph representation learning for DTI prediction.