Knowledge Graphs in Biomedical Information Science

Abstract

Biomedical knowledge graphs integrate heterogeneous entities including genes, diseases, drugs, proteins, and pathways into structured graph representations that enable reasoning, link prediction, and hypothesis generation across biological domains, yet adoption into biomedical research and clinical workflows varies across graph construction approaches and application contexts. We evaluated 216 biomedical knowledge graph programmes across centres in Italy and Austria between 2018 and 2022, spanning five category types: curated ontology-based knowledge graphs, literature-mined relation graphs, multi-omics-integrated biological networks, drug repurposing knowledge graphs, and clinical phenotype knowledge graphs. A Biomedical Knowledge Graph Index (BKGI) was constructed from five sub-scores -- entity coverage and completeness, relation accuracy, link prediction performance, drug repurposing or hypothesis generation yield, and query accessibility and API quality -- with weights from regression against sustained adoption. BKGI correlated with adoption at r = +0.84 and discriminated adopted from non-adopted graphs with an AUC of 0.884. Drug repurposing KGs scored highest (mean BKGI 0.826), while clinical phenotype KGs trailed at 0.600. Only 35.6 percent exceeded the 0.75 threshold. Entity coverage carried the largest weight (beta = +0.278), followed by link prediction performance (beta = +0.230).