DTI-DDI Fusion: A Unified Deep Learning Framework for Drug Interaction Prediction

Drug-Target Interaction (DTI) and Drug-Drug Interaction (DDI) are crucial problems to be solved. Traditional methods for DTI and DDI prediction are often costly and timeconsuming. The advancement of deep learning (DL) has shown the potent ability to solve this issue. The formal one affects the pace of discovering how drugs bind to the human body, the latter one influences the fitness of patients in clinical circumstances. By using DL technologies, DTI could concentrate on the interactions between drugs and targets, which helps developers detect the probability of drug binding to the human body. In this paper, we propose a novel method to combine DTI and DDI models by optimizing the DEEPScreen model for DTI prediction and the GGI-DDI model for DDI prediction. Our optimized DTI model leverages Convolutional Neural Networks (CNN) with residual blocks and masked attention mechanisms to improve prediction accuracy and feature extraction, reaching 0.9328 accuracy, 0.9536 Recall, and 0.9448 F1-Score. The optimized DDI model is based on GINE (GIN with Edge features), we replaced the attention mechanism and improved the structure of GINE to reduce the scale of parameters while maintaining high classification ability. Experiments on the ChEMBL dataset for DTI and the TWOSIDES dataset for DDI demonstrate that our optimized models outperform existing state-of-the-art methods in terms of accuracy, precision, recall, and other metrics. The combined DTIDDI framework provides a comprehensive solution for drug interaction analysis, assisting drug developers in screening potential drug candidates and helping clinicians optimize patient prescriptions to avoid adverse drug events.