A Knowledge-Guided Large Language Model Framework for Microbiome-Based Disease Diagnosis

Area of Science:

• Microbiome research
• Artificial intelligence in medicine
• Computational biology

Background:

• Gut microbiome analysis is promising for disease diagnosis but faces challenges like high dimensionality and small sample sizes.
• Traditional machine learning methods often overfit data and fail to capture true biological relationships, leading to inaccurate diagnoses.
• Integrating biological knowledge is crucial for robust microbiome-based diagnostics.

Purpose of the Study:

• To develop a novel two-phase, knowledge-guided large language model (LLM) framework for accurate microbiome-based disease diagnosis.
• To address limitations of traditional machine learning in handling high-dimensional microbiome data and incorporating biological expertise.
• To create a generalizable framework applicable to various microbiome-associated diseases.

Main Methods:

• A two-phase LLM framework was proposed, integrating biomedical expertise and in-context learning.
• Phase 1: LLM identified disease-associated microbial taxa and inferred biological relationships, reducing feature dimensionality.
• Phase 2: Few-shot prompting guided the LLM for disease outcome classification using acquired domain knowledge.

Main Results:

• The framework successfully identified disease-associated taxa and their relationships, reducing feature space dimensionality.
• The LLM framework achieved 73.91% accuracy in diagnosing inflammatory bowel disease (IBD), outperforming an optimized XGBoost classifier.
• Demonstrated superior performance and generalizability compared to traditional machine learning approaches.

Conclusions:

• The knowledge-guided LLM framework offers a powerful and generalizable strategy for microbiome-based disease diagnosis.
• This approach effectively leverages LLMs by integrating biomedical knowledge, overcoming limitations of conventional methods.
• Opens new avenues for disease diagnosis utilizing LLMs in the era of big data and artificial intelligence.