SDV-HGNN: similarity-based dual view heterogeneous graph neural network method for drug adverse side effect prediction

Background: Drug adverse side effects (ASEs) significantly impact public health, healthcare costs, and drug discovery processes. As medication usage increases, effective management of drug side effects becomes crucial. Previ- ous research has focused on single-perspective drug features such as chemical structure or topological information from knowledge graphs. Recent approaches attempt to learn separately from molecular graphs and drug-side effect net- works, combining these representations for prediction. However, these methods often report limited performance metrics and may not fully capture the complex interplay between molecular structures and broader drug-side effect relationships. Results: We propose a novel Similarity-based Dual View Heterogeneous Graph Neural Network (SDV-HGNN) for predicting drug adverse side effects. This approach simultaneously learns microscopic drug substructure features from the molecular graph and macroscopic features from a connectivity-enhanced Drug- adverse Side-effect Network (DSN). We introduced four additional edges between drugs and three between side effects using multiple context-specific similarity metrics. The problem is framed as a binary classification task within the context of link prediction on a graph. Our model demonstrated superior performance in 10-fold cross-validation (CV) using a benchmark dataset, achieving an AUROC of 0.8989 ± 0.0069, AUPR 0.9093 ± 0.0068, and F1 0.8261 ± 0.0056. The source code is available from GitHub at https://github.com/mayankkom-dev/ SDV-HGNN. Conclusions: The SDV-HGNN model shows promising results in predicting drug adverse side effects by leveraging both microscopic and macroscopic features simultaneously. By reporting a comprehensive set of performance metrics, our study provides a more thorough evaluation of the model’s capabilities, addressing previous research limitations.