Related Experiment Videos
HerbWise: A domain-specific large language model for traditional herbal medicine
Wenke Xiao1,2,3, Qian Cheng1,2, Yufei Mao2,4
1School of Pharmacy, Chengdu University of Traditional Chinese Medicine, Chengdu 611137, China.
Objective:
Traditional herbal medicine (THM) plays a vital role in global healthcare, contributing significantly to health maintenance, disease prevention, and chronic illness management. Despite its widespread use, the modernization of THM faces significant challenges, including the lack of standardized methodologies and insufficient scientific validation. To bridge these gaps, this study presents HerbWise, a large language model (LLM) specifically designed for THM research and applications.
Methods:
Built upon the ChatGLM3-6B framework, HerbWise integrates multidisciplinary knowledge from a high-quality dataset of 748 041 curated entries, encompassing herbal origins, medicinal parts, clinical applications, genomic data, and chemical constituents. HerbWise is trained using a multi-stage approach that combines pre-training, supervised fine-tuning (SFT), and reinforcement learning with human feedback (RLHF), enabling deep analysis and cross-disciplinary integration of THM knowledge. Additionally, HerbWise incorporates a retrieval-augmented generation (RAG) framework, enhancing its capabilities in compound interpretation, absorption, distribution, metabolism, excretion, and toxicity (ADMET) prediction, and knowledge-based question answering.
Results:
Comprehensive benchmarking and case studies demonstrate that HerbWise outperforms existing general-purpose and domain-specific models across multiple traditional herbal medicine tasks. It not only accurately answers complex domain-specific knowledge questions but also effectively supports compound property prediction, demonstrating robust knowledge integration and reasoning capabilities. These functions are delivered through a user-friendly web interface (https://cbcb.cdutcm.edu.cn/HerbWise/) to support real-world applications.
Conclusion:
By establishing a new standard for intelligent THM integration, this work underscores the transformative potential of LLMs in driving the scientific validation and global adoption of traditional medicine.
