Knowledge graph-based thought: a knowledge graph-enhanced LLM framework for pan-cancer question answering.

Area of Science:

• Biomedical Sciences
• Artificial Intelligence
• Computational Biology

Background:

• Large language models (LLMs) show potential in biomedical sciences but suffer from factual inaccuracies and hallucinations.
• Real-world applications of LLMs are hindered by unreliable outputs.

Purpose of the Study:

• To develop a novel framework, knowledge graph-based thought (KGT), that integrates LLMs with knowledge graphs (KGs).
• To improve the accuracy and reliability of LLM responses in the biomedical domain.

Main Methods:

• Integrated LLMs with knowledge graphs (KGs) to create the KGT framework.
• Utilized verifiable information from KGs to enhance LLM reasoning and reduce errors.
• Developed a pan-cancer question answering benchmark using a pan-cancer knowledge graph.

Main Results:

• The KGT framework significantly reduces factual errors in LLM reasoning.
• Demonstrated strong adaptability across various open-source LLMs.
• Facilitated drug repurposing discovery and prediction of drug resistance by analyzing cancer associations, biomarkers, and genetic mechanisms.

Conclusions:

• The KGT framework substantially enhances the accuracy and utility of LLMs in biomedical question answering.
• The study serves as a proof of concept for the framework's effectiveness in biomedical applications.