Background Identifying causative variants is crucial for the diagnosis of rare genetic diseases. Over the past two decades, the application of genome sequencing technologies in the field has significantly improved diagnostic outcomes. However, the complexity of data analysis and interpretation continues to limit the efficiency and accuracy of these applications. Various genotype and phenotype-driven filtering and prioritization strategies are used to generate a candidate list of variants for expert curation, with the final report variants determined through knowledge-intensive and labor-intensive expert review. Despite these efforts, the current methods fall short of meeting the growing demand for accurate and efficient diagnosis of rare disease. Recent developments in large language models (LLMs) suggest that LLMs possess the potential to augment or even supplant human labor in this context. Methods In this study, we have developed Genetic Transformer (GeneT), an innovative large language model (LLM) driven approach to accelerate identification of candidate causative variants for rare genetic disease. A comprehensive evaluation was conducted between the fine-tuned large language models and four phenotype-driven methods, including Xrare, Exomiser, PhenIX and PHIVE, alongside six pre-trained LLMs (Qwen1.5-0.5B, Qwen1.5-1.8B, Qwen1.5-4B, Mistral-7B, Meta-Llama-3-8B, Meta-Llama-3-70B). This evaluation focused on performance and hallucinations. Results Genetic Transformer (GeneT) as an innovative LLM-driven approach demonstrated outstanding performance on identification of candidate causative variants, identified the average number of candidate causative variants reduced from an average of 418 to 8, achieving recall rate of 99% in synthetic datasets. Application in real-world clinical setting demonstrated the potential for a 20-fold increase in processing speed, reducing the time required to analyze each sample from approximately 60 minutes to around 3 minutes. Concurrently, the recall rate has improved from 94.36% to 97.85%. An online analysis platform iGeneT was developed to integrate GeneT into the workflow of rare genetic disease analysis. Conclusion Our study represents the inaugural application of fine-tuned LLMs for identifying candidate causative variants, introducing GeneT as an innovative LLM-driven approach, demonstrating its superiority in both simulated data and real-world clinical setting. The study is unique in that it represents a paradigm shift in addressing the complexity of variant filtering and prioritization of whole exome or genome sequencing data, effectively resolving the challenge akin to finding a needle in a haystack.

The authors have declared no competing interest.

This study did not receive any funding

I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained.

Yes

The details of the IRB/oversight body that provided approval or exemption for the research described are given below:

Ethics committee/IRB of BGI gave ethical approval for this work.

I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals.

Yes

I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance).

Yes

I have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable.

Yes