Genetic Discovery Enabled by A Large Language Model.

Area of Science:

• Genomics
• Bioinformatics
• Artificial Intelligence in Medicine

Background:

• Large language models (LLMs) show promise for clinical applications in medicine.
• The potential of LLMs to accelerate genetic discovery remains largely unexplored.
• Investigating LLM utility requires robust experimental models, such as those from mouse genetics.

Approach:

• Utilized a specialized LLM, Med-PaLM 2, to analyze gene sets from murine models.
• Input free-text descriptions of biomedical traits and lists of high-impact alleles.
• Examined Med-PaLM 2's ability to identify causative genetic factors for various traits.

Key Points:

• Med-PaLM 2 accurately identified known causative murine genes for six biomedical traits, including diabetes and cataracts.
• The LLM successfully pinpointed a genetic factor for spontaneous hearing loss from comparative genomic analysis.
• A novel bigenic model for hearing loss susceptibility was developed based on the LLM's findings.

Conclusions:

• Med-PaLM 2 demonstrates capability in analyzing complex gene-phenotype relationships.
• LLMs can generate novel hypotheses, significantly facilitating the process of genetic discovery.
• This study validates the use of AI, specifically LLMs, as a tool for advancing genomic research.