DTI-MACF: Drug-Target Interaction Prediction via Multi-component Attention Network

Drug-target interaction (DTI) prediction plays an essential role in drug discovery. Traditional biomedical measurement via vitro experiments is reliable but can be prohibitively expensive, time-consuming, and inefficient, especially in large-scale datasets. In recent years, deep learning has been increasingly used in the biomedical field, especially for drug-target prediction. However, existing deep-learning-based DTI methods still need to improve in the aspect of feature extraction. In this paper, we propose a multi-component aggregation model with collaborative filtering for DTI prediction called DTI-MACF. Our approach constructs a bipartite graph to extract various potential features through multiple components module. To improve the accuracy of feature representation, we design a neighbourhood aggregator module based on the bipartite graph, which fuses abundant historical interactive information. We conduct extensive experiments on three benchmark datasets to demonstrate the strong competitiveness of our proposed model.