DrugReX: an explainable drug repurposing system powered by large language models and literature-based knowledge graph

TL;DR

DrugReX is a new system that uses large language models and a knowledge graph to repurpose drugs in a transparent and explainable way, improving trust in drug discovery.

Contribution

First integration of large language models for explainable drug repurposing using a literature-based knowledge graph.

Findings

DrugReX achieved high scores on 15 established drug repurposing cases.

Identified 25 candidate drugs for Alzheimer’s disease and related dementias, with nine clustering with FDA-approved drugs.

LLM-generated explanations were rated superior in quality and clarity compared to LLM-only explanations.

Abstract

Drug repurposing offers a time-efficient and cost-effective approach for therapeutic development by finding new uses for existing drugs. Additionally, achieving explainability in drug repurposing remains a challenge due to the lack of transparency in decision-making processes, hindering researchers’ understanding and trust in the generated insights. To address these issues, we present DrugReX, a system integrating a literature-based knowledge graph, embedding, scoring system, and explanation modules using large language models (LLMs). We validated DrugReX on 15 established drug repurposing cases, achieving significantly high scores. As a real-world use case, we applied DrugReX to identify candidate drugs for Alzheimer’s disease and related dementias (ADRD) and thoroughly evaluated the pipeline. The system identified 25 promising candidates, with nine clustering with FDA-approved ADRD drugs and 10 linked to ongoing clinical trials. For explainability, an LLM was employed to generate explanations supported by evidence from the literature-based knowledge graph. Domain expert evaluation revealed that DrugReX-produced explanations were superior in quality and clarity than using an LLM alone, enhancing the explainability of repurposing predictions. This study represents the first integration of LLMs to provide explainable insights into drug repurposing, bridging computation precision with explainability, and thus, ultimately enabling more transparent and reliable decision-making in therapeutic development.

Figures