Drug-Target Interaction Prediction Based on GCN and BiLSTM with a Cross-Attention Mechanism

Identifying drug-target interactions (DTI) is a critical aspect of drug discovery. While experimental methods are often costly and time-consuming, computational approaches are indispensable. In recent decades, deep learning has significantly advanced DTI prediction through techniques such as recurrent neural networks, convolutional neural networks, and graphical convolutional networks. However, these models have neglected detailed drug-target interactions, limiting their predictive power. In this study, we present an advanced deep learning (DL) model for DTI prediction. We integrate a graphical convolutional network (GCN) to represent drug molecules as graphs, capturing atomic interactions, and use a bidirectional short-term memory network (BiLSTM) to efficiently extract protein sequence features. We employ a cross-attention mechanism to enhance the capture of drug-target relationships and residual blocks to mitigate the vanishing gradient problem. Our model was compared to baseline models and advanced DL models using two standard datasets (C. elegans and human) to validate and assess its performance. The results indicate high performance. Our model achieved the highest recall (0.969) and performed competitively on other metrics (AUC = 0.947, Precision = 0.929, and F1-score = 0.949). A case study validated our model's efficacy, though some room for improvement remains.