Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From Knowledge to Treatment: Large Language Model Assisted Biomedical Concept Representation for Drug Repurposing

Chengrui Xiang, Tengfei Ma, Xiangzheng Fu, Yiping Liu, Bosheng Song, Xiangxiang Zeng

TL;DR

The paper tackles drug repurposing by enriching biomedical knowledge graphs with common-sense concept knowledge derived from large language models. LLaDR generates rich textual descriptions for entities and fuses them with KG embeddings through a dual-objective fine-tuning process that preserves graph structure while aligning with semantic text signals, formalized as $L = \zeta_1 L_{anc} + \zeta_2 L_{link}$ and involving a text-embedding alignment constraint. Empirical results on the DRKG dataset demonstrate state-of-the-art performance across multiple backbones and tasks, with extensive ablations showing the importance of prompt design, temperature, and semantic integrity. A Alzheimer’s disease case study highlights clinically plausible drug candidates, underscoring LLaDR’s potential impact on translational drug discovery, while acknowledging limitations related to LLM biases, scalability, and interpretability.

Abstract

Drug repurposing plays a critical role in accelerating treatment discovery, especially for complex and rare diseases. Biomedical knowledge graphs (KGs), which encode rich clinical associations, have been widely adopted to support this task. However, existing methods largely overlook common-sense biomedical concept knowledge in real-world labs, such as mechanistic priors indicating that certain drugs are fundamentally incompatible with specific treatments. To address this gap, we propose LLaDR, a Large Language Model-assisted framework for Drug Repurposing, which improves the representation of biomedical concepts within KGs. Specifically, we extract semantically enriched treatment-related textual representations of biomedical entities from large language models (LLMs) and use them to fine-tune knowledge graph embedding (KGE) models. By injecting treatment-relevant knowledge into KGE, LLaDR largely improves the representation of biomedical concepts, enhancing semantic understanding of under-studied or complex indications. Experiments based on benchmarks demonstrate that LLaDR achieves state-of-the-art performance across different scenarios, with case studies on Alzheimer's disease further confirming its robustness and effectiveness. Code is available at https://github.com/xiaomingaaa/LLaDR.

From Knowledge to Treatment: Large Language Model Assisted Biomedical Concept Representation for Drug Repurposing

TL;DR

The paper tackles drug repurposing by enriching biomedical knowledge graphs with common-sense concept knowledge derived from large language models. LLaDR generates rich textual descriptions for entities and fuses them with KG embeddings through a dual-objective fine-tuning process that preserves graph structure while aligning with semantic text signals, formalized as and involving a text-embedding alignment constraint. Empirical results on the DRKG dataset demonstrate state-of-the-art performance across multiple backbones and tasks, with extensive ablations showing the importance of prompt design, temperature, and semantic integrity. A Alzheimer’s disease case study highlights clinically plausible drug candidates, underscoring LLaDR’s potential impact on translational drug discovery, while acknowledging limitations related to LLM biases, scalability, and interpretability.

Abstract

Drug repurposing plays a critical role in accelerating treatment discovery, especially for complex and rare diseases. Biomedical knowledge graphs (KGs), which encode rich clinical associations, have been widely adopted to support this task. However, existing methods largely overlook common-sense biomedical concept knowledge in real-world labs, such as mechanistic priors indicating that certain drugs are fundamentally incompatible with specific treatments. To address this gap, we propose LLaDR, a Large Language Model-assisted framework for Drug Repurposing, which improves the representation of biomedical concepts within KGs. Specifically, we extract semantically enriched treatment-related textual representations of biomedical entities from large language models (LLMs) and use them to fine-tune knowledge graph embedding (KGE) models. By injecting treatment-relevant knowledge into KGE, LLaDR largely improves the representation of biomedical concepts, enhancing semantic understanding of under-studied or complex indications. Experiments based on benchmarks demonstrate that LLaDR achieves state-of-the-art performance across different scenarios, with case studies on Alzheimer's disease further confirming its robustness and effectiveness. Code is available at https://github.com/xiaomingaaa/LLaDR.
Paper Structure (19 sections, 6 equations, 4 figures, 10 tables)

This paper contains 19 sections, 6 equations, 4 figures, 10 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Drug Repurposing
  4. LLM incorporated Knowledge Representation
  5. LLaDR Framework
  6. Generating Discription
  7. Knowledge Graph Fine-Tuning
  8. Experimental Setup
  9. Experimental Results
  10. Case Analysis
  11. Conclusion
  12. Appendix
  13. Datasets
  14. Hardware
  15. Drug Repurposing Results
  16. ...and 4 more sections

Figures (4)

  • Figure 1: Comparison of standard KG embedding (top) and LLaDR (bottom). LLaDR incorporating semantic concept knowledge generates more meaningful representations, leading to better separation of entities.
  • Figure 2: Overview of LLaDR. Input and Output are highlighted at each step. Step 1: Obtain text embeddings for all entities in DRKG, achieved by merging word embeddings with discription embeddings retrieved from LLMs. Step 2: We utilize the initial vectors generated by LLMs and apply them to the training of knowledge graphs, thereby obtaining the final embeddings of entities and relations enriched with comprehensive semantics.
  • Figure 3: Robustness of LLaDR on noisy KGs by deleting or adding unknown associations.
  • Figure 4: The mechanism of action of Dasatinib and Quercetin in Alzheimer's disease.