Adverse drug reactions (ADRs) are statistically characterized within randomized clinical trials or by postmarketing pharmacovigilance. However, the molecular mechanisms causing ADRs remain unknown in most cases. This is true even for common toxicities that are classically monitored during trials such as hepatic or skin toxicities. Interestingly, many elements of knowledge about drugs and drug ingredients are available beside clinical trials. In particular, open-access knowledge graphs describe their properties, interactions, and involvements in pathways. Expert classifications have also been manually established by experts and label drugs either as causative or not for several types of ADRs. In our paper, we propose to mine biomedical knowledge graphs to identify biomolecular features that enable to automatically reproduce such expert classifications, distinguishing drugs causative or not for a given type of ADR. In an Explainable AI perspective, we explore simple classification techniques such as Decision Trees and Classification Rules because they provide human-readable models which explain the classification itself. We also evaluate the assumption that biomolecular features mined from knowledge graphs might provide elements of explanation for the molecular mechanisms behind ADRs.