LLM-Integrated Representative Path Selection for Context-Aware Drug Repurposing on Biomedical Knowledge Graphs

Haerin Song, Dongmin Bang, Bonil Koo, Sun Kim, Sangseon Lee

Keywords: Drug Repurposing, Biomedical Knowledge Graph, Large Language Model, Mechanism of Action, Path Selection, Biomedical Reasoning, Interpretability

Abstract: Drug repurposing, which seeks novel drug–disease associations by integrating biomedical knowledge, is hindered by modeling complex multi-hop relationships in knowledge graphs. We propose DrugCORpath, which integrates biomedical knowledge graphs with pretrained biomedical large language models (LLMs). Unlike node-centric methods, it captures biological context by converting multi-hop drug–disease paths into sentences reflecting plausible mechanisms of action (MoAs) and embedding them. We then apply clustering–based selective filtering with a distance metric to retain meaningful paths while removing redundancy and noise. Experiments show DrugCORpath outperforms graph-based, LLM-based, and path-based baselines in drug repurposing, achieving up to 4.9% higher accuracy than the prior SOTA. Analyses confirm that filtering reduces noise and enhances biological diversity, and case studies validate clinically relevant rationales, improving interpretability. Collectively, these results underscore the method’s potential for interpretable, biologically plausible drug repurposing.

Submission Number: 39