An interpretable machine learning framework for adverse drug reaction prediction from drug-target interactions

Background: Adverse drug reactions (ADRs) present challenges to patient safety and healthcare systems. Current pharmacovigilance methods, such as the Yellow Card Scheme (YCS), provide valuable post-marketing data, but the mechanistic causes of these ADRs are not fully understood. Leveraging drug-target interaction data with interpretable machine learning offers a promising approach to anticipate ADRs and understand their underlying mechanisms. Objective: This study proposes an interpretable machine learning (ML) framework to predict significant ADRs using drug-target interaction data. The framework aims to identify key pharmacological relationships, helping to inform drug safety. Methods: Drug-target interaction data from STITCH was combined with ADR reports from the YCS. Disproportionality analysis identified significant ADR signals which were used to train Random Forest classifiers across System Organ Class (SOC) categories. Class imbalance was addressed with SMOTE and Tomek, and Bayesian optimisation refined hyperparameters. Feature importance scores provided interpretability, and the top features were validated using known target-disease associations from DisGeNET. Results: Prediction performance varied across SOC categories, with ROC AUC scores up to 0.94. Feature importance analysis identified pharmacologically relevant targets, validated using DisGeNET and comparisons with SIDER highlighted the added value of real-world data. Conclusions: The interpretable ML framework links drug-target interactions to ADRs, offering a promising approach for predictive pharmacovigilance (PPV) and supporting safer drug development.