Introduction:
The field of medical genetics has witnessed rapid advancements, primarily driven by the increased availability of genomic sequencing and big data analytics. This has expanded our understanding of genetic diseases and heightened the demand for clinical geneticists, who play a pivotal role in diagnosing and managing genetic disorders.

Natural Language Processing (NLP), particularly through large language models (LLMs) like GPT, is emerging as a transformative tool in medicine. Trained on vast amounts of text data, these advanced artificial intelligence systems are designed to understand and generate human language, in order to perform various tasks such as answering questions, generating text, summarizing information, and more. Despite their promise, LLMs face multiple challenges.

To understand the potential and limitations of LLMs in medical genetics, this review follows the typical steps performed by medical geneticists in the clinic. We highlight potential applications of LLMs at each step, review relevant studies published so far, and emphasize key challenges that remain unresolved.

Methods:
We searched PubMed for all original articles published in 2023-2024 that study the applications of LLMs in medical genetics. In total 15 relevant papers were found and discussed throughout our review.

Results:
Following the typical workflow of a medical geneticist, we demonstrate that LLMs have the potential to contribute at each step; Starting with identifying individuals within electronic medical records (EMRs) who may benefit from genetic counseling, normalization of clinical phenotype terms in unstructured clinical notes, and predictive modeling of the likelihood of specific rare genetic conditions based on clinical presentation. Then, LLMs can be used for data integration to support comprehensive differential diagnosis. They can also assist in results interpretation, such as improving variant classification and gene prioritization. Even after a genetic diagnosis is made, LLMs could assist in analyzing the literature to educate the patient and family, and aid in long-term follow-up management by automating appointments, monitoring symptoms, and tailoring personalized treatment options. Lastly, LLMs are increasingly useful for research purposes, including identifying therapeutic targets, selecting individuals for clinical trials, analyzing statistics, and refining scientific writing.

However, while current research indicates that LLMs can outperform traditional NLP methods and, in some cases, achieve performance equivalent to humans, these models also exhibit inconsistencies and are prone to generating plausible yet incorrect responses. Furthermore, LLMs rely heavily on available databases, which can lead to biased responses, particularly concerning underrepresented populations or rare genetic findings.

Patient communication remains an area in need of improvement, especially in medical genetics where sensitive results impacting the patient and their family are reported, underscoring the importance of human interaction for providing compassionate genetic counseling and support. Ethical considerations add another layer of complexity, with issues related to privacy, data security, and informed consent being paramount when integrating AI technologies into healthcare.

Conclusion:
The revolution of LLMs in our daily lives and medical practice has already begun and is unstoppable. It is essential to embrace and leverage this technology for the benefit of both patients and physicians. LLMs indeed hold great promise for transforming various aspects of medical genetics; however, their integration must be approached thoughtfully. Ongoing research, validation, and adherence to ethical standards are essential to ensure that these technologies serve as complementary tools, enhancing rather than replacing the invaluable role of human expertise in genetics.