Integrating Multi-Source Biomedical Data for Adverse Drug Reaction Analysis: A Knowledge Graph Approach

Understanding adverse drug reaction mechanisms requires integrating information from multiple heterogeneous sources. We developed a pipeline for combining drug-target data (DrugBank), biochemical pathways (Reactome), and adverse drug reactions identified through Natural Language Processing methods applied on publicly available documents and package inserts (SIDER) into unified RDF/OWL knowledge graphs. The pipeline extracts: all protein targets for a drug from DrugBank, all biochemical reactions involving the drug from Reactome, and the top five reported adverse reactions from SIDER. Applied to metformin as proof-of-concept, the resulting graph contains 139 triples linking 1 drug, 3 protein targets, 2 biochemical reactions, 2 pathways, and 5 adverse reactions. This demonstrates technical feasibility of multi-source integration, with automation as next step.