IMAEN: An interpretable molecular augmentation model for drug–target interaction prediction

Abstract

Drug discovery is a crucial aspect of biomedical research, and predicting drug–target interactions (DTIs) is a vital step in this process. Graph neural networks (GNNs) have achieved remarkable performance on graph learning tasks, including DTI prediction. However, existing DTI models using GNNs are insufficient to aggregate node neighborhood information and lack interpretability. In this study, we propose an interpretable molecular augmentation model named IMAEN for drug–target interaction prediction, which employs molecular augmentation mechanism to fully aggregate the molecular node neighborhood information, especially for the nodes with fewer neighborhoods. Moreover, we design an interpretable stack convolutional encoding module to process protein sequence from the perspective of multi-scale and multi-level interpretably. Compared with the existing models, our proposed model has the best effect and achieves the best performance on four benchmark datasets. The visualization of the model and interpretation of its predictions provide valuable insights into the underlying mechanisms of drug–target interactions, which could assist researchers in narrowing the search space of binding sites and aid in the discovery of new drugs and targets. The source code and experimental datasets can be found in https://github.com/zhangjing-dmu/IMAEN.