Context-Aware Safe Medication Recommendations with Molecular Graph and DDI Graph Embedding

Molecular structures and Drug-Drug Interactions (DDI) are recognized as important knowledge to guide medication recommendation (MR) tasks, medical concept embedding has been applied boost their performance. Though promising performance achieved by leveraging Graph Neural Network (GNN) models encode the molecular of medications or/and DDI, we observe that existing still defective: 1) differentiate with similar molecules but different functionality; 2) properly capture unintended reactions between drugs in space. To alleviate this limitation, propose Carmen, a cautiously designed graph embedding-based MR framework. Carmen consists four components, including patient representation learning, context information extraction, context-aware GNN, DDI encoding. incorporates visit history into learning graphs distinguish topology dissimilar activity. Its encoding module is specially devised for non-transitive interaction graphs. The experiments on real-world datasets demonstrate achieves remarkable improvement over state-of-the-art can improve safety recommended proper