Polypharmacy, the concurrent use of multiple medications, poses significant risks due to potential adverse drug events (ADEs) arising from drug-drug interactions (DDIs). These ADEs, stemming from complex combinations of drugs rather than individual drugs, pose challenges for prediction and management. This unpredictability complicates patient care and increases risks for healthcare providers, highlighting the need for accurate prediction systems with clear explanations. Methodology. We propose a novel approach based on reinforcement learning using a modified Hetionet, a comprehensive biomedical knowledge graph from 29 databases. Our model employs multi-hop reasoning to explore complex drug interactions, enhancing the interpretability of predicted DDIs. By modifying Hetionet, we gained deeper insights into drug relationships and their potential contribution to specific ADEs. The RL framework enables our model to navigate the knowledge graph, uncovering hidden connections among drug pairs leading to harmful interactions. Results. The results demonstrate the efficacy of our model in predicting unknown DDIs, supported by extensive ablation tests highlighting the significant contributions of various model components. Additionally, our model provides explainable evidence through traversal paths, aiding clinicians in understanding the rationale behind risky drug combinations and facilitating safer medication selection. Furthermore, our approach yields relevant biological explanations for observed interactions, indicating its potential to enhance safety in polypharmacy management. Conclusion and Discussion: Our research highlights the importance of accurate prediction and explainability in mitigating risks associated with polypharmacy. By leveraging RL on knowledge graphs, our approach offers a promising solution for improving patient safety and healthcare provider decision-making in managing complex drug interactions. Further research could explore the scalability and generalization ability of our approach across diverse healthcare settings and patient populations.