CL-MHAD: Contrastive Learning-based Multi-Hypergraph Aggregation and Diffusion model for prescription recommendation.

Area of Science:

• Artificial Intelligence
• Computational Biology
• Traditional Chinese Medicine (TCM)

Background:

• Personalized diagnosis and treatment in TCM rely on syndrome-based prescription recommendations.
• Extracting and fusing multi-dimensional herb knowledge for accurate TCM recommendations is challenging.

Purpose of the Study:

• To propose CL-MHAD, a contrastive learning-based multi-hypergraph aggregation and diffusion model for improved TCM prescription recommendations.
• To address the challenge of effectively extracting and fusing multi-dimensional herb knowledge.

Main Methods:

• Developed a multi-view hypergraph reconstruction mechanism focusing on prescriptions, herbs, properties, and dosages.
• Implemented a diffusion-enhanced method for capturing high-order relationships and a cross-view contrastive learning strategy.
• Utilized topology-aware random walk augmentation to mitigate data sparsity and an integrated loss function for optimization.

Main Results:

• CL-MHAD demonstrated superior performance over baseline models in a real-world clinical setting for Gastrointestinal Diseases (GID).
• Achieved performance gains ranging from 1.27% to 24.67% across multiple evaluation metrics.
• Validated the effectiveness of the weighted fusion strategy and the model's robustness against data sparsity.

Conclusions:

• CL-MHAD offers an effective solution for accurate syndrome-based prescription recommendations in TCM.
• The proposed model presents a promising paradigm for advancing personalized medicine in TCM.