Enhancing gene set overrepresentation analysis with large language models.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Traditional gene set overrepresentation analysis (ORA) relies on static, human-curated databases, limiting flexibility in interpreting high-throughput transcriptomics and proteomics data.
• Existing methods struggle to adapt to specific biological contexts or dynamically generated gene lists.

Purpose of the Study:

• To develop a flexible framework, llm2geneset, that utilizes large language models (LLMs) to dynamically generate gene set databases.
• To enable context-aware functional interpretation of biological data by integrating LLM-generated gene sets with analysis methods like ORA.
• To benchmark the performance of LLM-generated gene sets against human-curated databases.

Main Methods:

• Development of the llm2geneset framework, leveraging LLMs to create gene sets based on input genes and natural language biological context.
• Integration of dynamically generated gene sets with established analysis methods, such as ORA, for functional annotation.
• Comparative analysis of LLM-generated gene sets against human-curated databases using benchmarking studies.
• Application of the framework to RNA-sequencing data from iPSC-derived microglia treated with a TREM2 agonist.

Main Results:

• LLM-generated gene sets demonstrated comparable quality to human-curated gene sets.
• The llm2geneset framework successfully identified biological processes within input gene sets, outperforming traditional ORA and direct LLM prompting.
• The framework facilitated flexible, context-aware gene set generation and improved the interpretation of high-throughput biological data, as shown in the TREM2 agonist study.

Conclusions:

• llm2geneset provides a powerful and flexible alternative to traditional gene set enrichment analysis, utilizing LLMs for dynamic database generation.
• The framework enhances the interpretation of complex biological datasets by offering context-specific functional annotations.
• llm2geneset represents a significant advancement in bioinformatics tools for biological data analysis and discovery.