RepurposeAgent: Large Language Models Meet Biomedical Knowledge Graphs for Mechanistically Grounded Therapeutic Prioritization

Overview

RepurposeAgent is a disease-centric framework that integrates biomedical knowledge graphs (KGs) with large language models (LLMs) to enable systematic drug repurposing and disease-specific drug ranking.

The framework combines:

• Structured knowledge (drug–gene–disease relations)
• Unstructured evidence (biomedical literature)
• Perturbational transcriptomics (LINCS)
• LLM-based reasoning and summarization

Unlike conventional drug-repurposing pipelines that focus on binary drug–disease prediction, RepurposeAgent provides an interpretable, multi-evidence ranking of candidate drugs, with optional mechanistic explanations and disease subtype–aware analysis.

---

Key Contributions

• Unified KG + LLM framework for drug repurposing
• Disease-specific drug ranking instead of binary association prediction
• Multi-source evidence integration (KG, literature, LINCS, GSEA)
• Mechanistic hypothesis generation grounded in biological pathways
• End-to-end automated pipeline

---

System Pipeline

Disease Query
   ↓
Disease Standardization
   ↓
Candidate Drug Generation and Evidence Integration
   ↓
Disease- and Drug-related Gene Evidence Integration
   ↓
Pathway Perturbation Analysis (LINCS / GSEA)
   ↓
Evidence-grounded Confidence Scoring and Drug Ranking

---

Repository Structure

RepurposeAgent/
├── RepurposeAgent.sh              # End-to-end pipeline
├── input/                         # Disease input files
├── output/                        # Generated results
├── prompts/                       # LLM prompts
├── DB/                            # Knowledge bases and omics data
├── MedCPT.npy/                    # Disease embeddings
├── requirements.txt
└── README.md

---

Installation

Requirements

• Python ≥ 3.11
• Conda (recommended)
• Java (required for GSEA)
• Download files and store in RepurposeAgent folder

Setup

git clone https://github.com/your-org/RepurposeAgent.git
cd RepurposeAgent

conda create -n RepurposeAgent python=3.11
conda activate RepurposeAgent

pip install -r requirements.txt

---

LLM Configuration

RepurposeAgent uses OpenAI-compatible LLMs.

Create a parameter file (not committed to GitHub):

parameter.gpt51.txt

Example:

OPENAI_API_KEY=your_api_key
MODEL=gpt-5.1

---

Usage

Run the complete pipeline for a disease:

bash RepurposeAgent.sh "Acute Myeloid Leukemia"

---

Output

Main output file:

output/<Disease>.final_prediction.tsv

---

Citation

If you use RepurposeAgent, please cite:

@article{RepurposeAgent,
  title={Large Language Models Meet Biomedical Knowledge Graphs for Mechanistically Grounded Therapeutic Prioritization},
  author={Wei, Chih-Hsuan and Day, Chi-Ping and Wang, Zhizheng and others},
  journal={TBD},
  year={2026}
}

---

Disclaimer

RepurposeAgent is intended for research and hypothesis generation only and is not designed for direct clinical decision-making. RepurposeAgent shows the results of research conducted in the Computational Biology Branch, DIR/NLM. The information produced on this website is not intended for direct diagnostic use or medical decision-making without review and oversight by a clinical professional. Individuals should not change their health behavior solely on the basis of information produced on this website. NIH does not independently verify the validity or utility of the information produced by this tool. If you have questions about the information produced on this website, please see a health care professional.

---

Contact

For questions or collaborations, please contact the author, Chih-Hsuan Wei (chih-hsuan.wei@nih.gov).