Drug-Drug Interaction Event Prediction Based on Multi-Relational Enhanced Graph Neural Network

Drug-drug interaction (DDI) events can lead to unintended adverse consequences, and their prediction is an important part of medication safety. In recent years, models based on Graph Neural Networks (GNNs) have made significant progress in this area, but existing methods tend to utilize only drug structure information or interaction information. In addition, due to the scarcity of labeled instances of rare events, predicting these events faces challenges. In this study, we propose a novel graph neural network model based on Multi-relational Contrastive Learning Graph Neural Network (MRCGNN), Multi-Relational Enhanced Graph Neural Network (MRE-GNN), which can incorporate drug structure information and a multi-relational contrastive learning strategy to capture the implicit features of rare DDI events. By deploying GNN on the multi-relational DDI event graph, the model extracts the structural features extracted from the drug molecular graph and further enhances the capture ability by multi-view negative correspondence enhancement strategies (such as random permutation of nodes and edges, introduction of noise, adding or removing nodes or edges, or both). Finally, the model combines the drug structural features with the drug pair representation to predict DDI events. Both on Deng's and Ryu’s datasets, MRE-GNN performs better than the single eigenvector model and shows satisfactory results in the prediction of rare DDI events.