Graph-Based Personalized Medication Recommendation Using EHR and Drug Molecular Representations

The provision of intelligent drug recommendations is beneficial in facilitating accurate medication and reducing adverse reactions. However, challenges remain, including inadequate analysis of patient data and inaccurate modeling of drug interactions. To address these, we propose FG-GRN (Fused Graph-Gated Recurrent Network), a personalized drug recommendation framework that integrates EHRs and 3D drug molecular graphs. Our key innovations include: (1) a unified framework fusing temporal EHR features with spatial drug structures; (2) a dynamic noise suppression mechanism for long-term patient data; and (3) an interpretable patient-specific drug influence matrix for transparency. Experiments on 61,267 patients show FG-GRN achieves a PRAUC of 74.77.