A Multimodal Data Fusion-Based Deep Learning Approach for Drug-Drug Interaction Prediction

Prediction of drug-drug interaction (DDI) is one of the vital topics in drug development. Many computational methods have been present for DDI prediction. However, these methods are often limited to exploiting the drug’s molecular structure and ignoring other features of modalities, necessary for capturing the complicated DDI patterns. In this study, we proposed HF-DDI, a hybrid fusion-based deep learning framework, for DDI event prediction using various biomedical information about drugs. At first, HF-DDI uses multiple drug similarities based on drug substructure, target, and enzyme as representations of drug-drug interaction events. Afterward, HF-DDI combines two different levels of fusion strategies and utilizes a score calculation module with adaptive weighted averaging to help prediction-making. Experimental results demonstrated that our proposed method outperformed existing approaches for interaction prediction, which provided a high accuracy of 0.948.