Genetic Diagnosis and Discovery Enabled by Large Language Models

N/A (LLMs for rare genetics)

Evaluation of Gemini 2.5 Pro and Med-PaLM 2 for solving genetics problems of increasing complexity: Gemini identifies causal genetic factors in 6 human patients without fine-tuning; Med-PaLM 2 discovers a new genetic factor for deafness validated in vivo by knock-in mouse

First peer-reviewed article documenting the capacity of LLMs to contribute to diagnostic discovery in rare genetics beyond text synthesis. Gemini 2.5 Pro identifies causal genetic factors in 6 rare disease patients without task-specific fine-tuning. Med-PaLM 2 correctly identifies causal genes in six mouse models and discovers a new genetic factor for spontaneous deafness, experimentally validated by knock-in mouse. Hallucinations are still reported — medical supervision is essential.

Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.

A milestone: LLMs no longer just summarize literature, they actively contribute to diagnostic reasoning. The functional validation of the candidate identified by Med-PaLM 2 (knock-in mouse) elevates this article above a chatbot benchmark. In clinical practice, LLMs can be used for reviewing unresolved cases with candidate variant lists — always with expert medical supervision.

first peer-reviewed article validating LLM diagnostic capacity in rare genetics +2; functional knock-in mouse validation (Med-PaLM 2) +2; Gemini 2.5 Pro without fine-tuning +2; medical supervision documented as necessary +1