A deep neural network model for optimizing traditional Chinese medicine prescriptions with data augmentation

Area of Science:

• Artificial Intelligence in Medicine
• Computational Pharmacology
• Traditional Chinese Medicine Research

Background:

• Traditional Chinese Medicine (TCM) prescription optimization faces challenges with noisy data and excessive modifications.
• Existing clinical decision support systems inadequately address these critical issues in TCM.

Purpose of the Study:

• To develop a deep learning framework, DA-TCMPO, for optimizing TCM prescriptions.
• To tackle data noise and minimize risks from prescription modifications in TCM.

Main Methods:

• Proposed DA-TCMPO framework utilizing tailored data augmentation techniques.
• Incorporated a Diffusion Model Based on Double Attention (DAD) for sample diversity and a Variable Noise Embedding Module (VNE) for denoising.
• Developed the Chinese Herbal Prescriptions for Diseases (CH) dataset for model training.

Main Results:

• DA-TCMPO significantly outperformed baseline models on the CH dataset, with precision, accuracy, recall, and F1-score improvements of 67.8%, 83.3%, 83.3%, and 83.6%, respectively.
• In vivo validation in Ulcerative Colitis (UC) mouse models showed DA-TCMPO-optimized prescriptions (e.g., CYKKL-2) significantly improved key health indicators.
• Demonstrated practical efficacy of optimized TCM prescriptions in a preclinical model.

Conclusions:

• DA-TCMPO shows promise for enhancing diagnostic and therapeutic decision-making in TCM.
• The framework offers a viable solution for optimizing TCM prescriptions in clinical practice.
• Highlights the potential of deep learning and data augmentation in advancing TCM research.