[3S5-OS-7c] OS-7

[3S5-OS-7c-01] JGCLLM: A Japanese Genetic Counseling Large Language Model

〇Takuya Fukushima¹, Masae Manabe¹, Shuntaro Yada¹, Shoko Wakamiya¹, Eiji Aramaki¹, Akiko Yoshida^3,4, Yusaku Urakawa^3,5, Akiko Maeda³, Shigeyuki Kan², Masayo Takahashi² (1. Nara Institute of Science and Technology, 2. Vision Care Inc., 3. Kobe City Eye Hospital, 4. Center for Genomic Medicine, Graduate School of Medicine, Kyoto University, 5. Department of Medical Oncology, Kobe City Medical Center General Hospital)

Keywords:Medical Natural Language Processing, Natural Language Processing, Large Language Model, Genetic Counseling, Chatbot

Advances in genetics research and treatment have increased the demand for genetic counseling. However, genetic counseling requires specialized medical knowledge and counseling skills, and the high cost of education has led to a shortage of specialists. In recent years, with the rapid development of Large Language Models (LLMs), expert-level competence has been reported in various fields, and it is expected that LLMs will be utilized in genetic counseling. To construct JGCLLM, we applied LLM improvement methods (Instruction Tuning, RAG, Prompt Engineering) to data collected from the Web and data generated with experts. For the evaluation of JGCLLM, 120 questions were selected from those collected through crowdsourcing. Genetic counselors evaluated JGCLLM's responses to these questions to determine the impact and challenges of each improvement technique on LLM in genetic counseling.