Large language models identify causal genes in complex trait GWAS.

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Identifying causal genes at genome-wide association study (GWAS) loci is crucial for understanding complex traits but remains a significant challenge.
• Current literature-mining methods often lack the accuracy and scalability needed for comprehensive genetic analysis.

Purpose of the Study:

• To evaluate the effectiveness of large language models (LLMs) in prioritizing likely causal genes at GWAS loci.
• To compare LLM performance against existing state-of-the-art methods and assess their generalizability to novel loci.

Main Methods:

• Systematic evaluation of general-purpose LLMs using benchmark datasets of high-confidence causal genes.
• Inclusion of a unique dataset from 23 unpublished GWAS to test performance on novel loci.
• Assessment of LLM performance when integrated with existing genetic analysis methods.

Main Results:

• LLMs demonstrated high accuracy in prioritizing causal genes at GWAS loci, outperforming or matching current state-of-the-art methods.
• LLMs showed robust performance on novel loci, indicating strong generalizability.
• Integrating LLMs with existing methods significantly enhanced overall causal gene identification performance.

Conclusions:

• LLMs provide an accurate, scalable, and generalizable approach for causal gene identification in GWAS.
• This work establishes LLMs as a powerful tool to accelerate the discovery of genes underlying complex traits.
• LLMs represent a significant advancement in leveraging artificial intelligence for genetic research.