Inferring Drug-Gene Relationships in Cancer Using Literature-Augmented Large Language Models

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Extracting drug-gene relationships from literature is crucial for cancer therapy.
• The volume of biomedical literature presents challenges for manual extraction.

Purpose of the Study:

• To develop an automated pipeline for inferring drug-gene interactions using retrieval-augmented large language models (LLMs).
• To address limitations of static LLMs, such as outdated knowledge and potential inaccuracies.
• To create a user-friendly tool for cancer researchers to explore drug-gene relationships.

Main Methods:

• Integrated PubMed and state-of-the-art LLMs for literature analysis.
• Developed a retrieval-augmented LLM pipeline to infer drug-gene interactions.
• Validated pipeline performance using curated databases and constructed a pan-cancer drug-gene network.
• Created GeneRxGPT, a web application for accessible use.

Main Results:

• The pipeline accurately identifies established and emerging drug targets.
• A pan-cancer drug-gene interaction network was constructed.
• Identified a novel association between CTNNB1 mutations and sorafenib sensitivity in liver cancer.

Conclusions:

• The developed pipeline provides accurate, evidence-based drug-gene inferences from up-to-date literature.
• GeneRxGPT facilitates exploration of drug-gene relationships, accelerating drug discovery.
• This approach empowers researchers to advance targeted cancer therapies and drug repurposing.