A Survey on Unifying Large Language Models and Knowledge Graphs for Biomedicine and Healthcare.

Area of Science:

• Artificial Intelligence in Biomedicine and Healthcare
• Knowledge Representation and Reasoning
• Digital Health

Background:

• Large language models (LLMs) show promise in AI but struggle with factual accuracy and reasoning in critical domains like healthcare.
• Biomedical data is vast, complex, multimodal, and often siloed, hindering unified AI analysis.
• Existing AI models face challenges in accuracy, controllability, and interpretability, crucial for biomedical applications.

Purpose of the Study:

• To systematically investigate and summarize recent studies on unifying LLMs and KGs.
• To explore how this unification can leverage complex biomedical data.
• To accelerate the development of next-generation AI for biomedicine and healthcare.

Main Methods:

• Systematic review of recent literature on LLM and KG integration.
• Analysis of approaches to enhance LLM reasoning and knowledge grounding.
• Exploration of LLM generalizability across diverse biomedical data modalities.

Main Results:

• Emerging research focuses on empowering LLMs with planning, reasoning, and knowledge grounding capabilities.
• LLMs show potential in automating the processing of complex healthcare data into unified knowledge graphs (KGs).
• The unification of LLMs and KGs offers a pathway to overcome data integration challenges.

Conclusions:

• Unifying LLMs and KGs is crucial for unlocking the full potential of AI in biomedicine and healthcare.
• This integration can enhance the reliability, accuracy, and interpretability of AI models.
• Further research in this area is vital for advancing digital health and AI-driven medical applications.