A survey for large language models in biomedicine

Area of Science:

• Artificial Intelligence
• Biomedical Informatics
• Natural Language Processing

Background:

• Large language models (LLMs) demonstrate advanced capabilities in natural language understanding and generation.
• Existing biomedical LLM surveys often lack comprehensive analysis across diverse applications and recent advancements.
• This review focuses on the practical implications of LLMs in real-world biomedical contexts.

Purpose of the Study:

• To provide an in-depth examination of the current landscape, applications, challenges, and prospects of LLMs in biomedicine.
• To analyze LLM capabilities in zero-shot learning and adaptation strategies for specialized biomedical tasks.
• To identify future research directions for responsible and effective LLM deployment in healthcare.

Main Methods:

• Systematic review of 484 publications from PubMed, Web of Science, and arXiv.
• Analysis of LLM applications in areas such as diagnostic assistance, drug discovery, and personalized medicine (137 studies).
• Examination of LLM adaptation strategies, including fine-tuning for uni-modal and multi-modal models.

Main Results:

• LLMs exhibit significant potential in zero-shot learning across various biomedical tasks.
• Fine-tuning strategies are crucial for enhancing LLM performance in specialized areas like medical question answering.
• Key challenges include data privacy, model interpretability, dataset quality, and ethical considerations.

Conclusions:

• LLMs offer transformative potential in biomedicine, but challenges must be addressed for safe and effective implementation.
• Future research should focus on privacy-preserving methods like federated learning and explainable AI.
• Continued development is essential to harness LLM capabilities responsibly in healthcare.