Graph-based deep learning for drug-drug interaction prediction: a systematic review

Identifying potential drug-drug interactions (DDIs) is crucial for drug development and therapeutic safety, motivating increasing efforts in computational DDI prediction. Although several surveys have summarized recent advances, a systematic review that explicitly organizes existing studies from the perspective of graph-based deep learning paradigms is still lacking. In this work, we present a comprehensive review of graph-based methods for DDI prediction, with a particular focus on three representative technical routes: graph convolutional networks (GCNs), graph attention networks (GATs), and graph contrastive learning (GCL). We review and categorize representative DDI prediction models according to these three paradigms, highlighting their modeling strategies, advantages, and limitations in capturing molecular structures, heterogeneous interactions, and robust representations. We then discuss key challenges and future research directions, emphasizing multi-modal data integration and model interpretability. This review aims to provide a structured overview of the current landscape and to serve as a practical reference for the development of more accurate, robust, and interpretable DDI prediction models.