<p>Herb–drug interactions (HDIs) have emerged as a growing concern in modern pharmacotherapy, particularly with the increasing use of herbal medicines and natural dietary supplements alongside conventional drugs. These interactions can compromise drug efficacy or safety by altering pharmacokinetic or pharmacodynamic processes. Despite their clinical importance, HDIs remain insufficiently studied, largely due to the complexity and cost of experimental validation. In this study, we propose a novel graph-based computational framework for predicting the effects of HDIs. Our approach leverages a heterogeneous, multi-relational graph that integrates drug–drug interactions (DDIs), drug–target interactions (DTIs), and herb–target interactions (HTIs). To capture the semantic richness of this complex graph, we employ metapath-guided random walks to generate node embeddings, which are then combined with molecular fingerprints, tabular descriptors, and target node encodings to train various supervised classifiers. Extensive experiments confirm the effectiveness of our approach, especially with tree-based models such as XGBoost and Random Forest, which achieve high predictive performance in multi-class classification tasks. These results underscore the potential of AI-powered, graph-based methodologies to support integrative and precision medicine by enabling scalable and accurate prediction of herb–drug interaction effects.</p>