A Feature-Aware Approach to Acupoint Compatibility Prediction Using Residual Graph Attention Networks and Matrix...

Area of Science:

• Traditional Chinese Medicine
• Computational Biology
• Bioinformatics

Background:

• Acupoint compatibility is crucial for effective acupuncture treatment.
• Existing link prediction models may overlook acupoint features and be susceptible to noise.
• Data-driven approaches are needed to explore and validate acupoint compatibilities.

Purpose of the Study:

• To propose a novel model for predicting acupoint compatibility.
• To address limitations of existing methods by integrating acupoint features and structural information.
• To enhance the accuracy and scope of acupuncture treatment recommendations.

Main Methods:

• Developed a Feature-Aware Residual Graph Attention Network and Matrix Factorization (FRGATMF) model.
• Implemented a feature-aware connectivity fusion strategy to enrich acupoint representations.
• Utilized deep non-negative matrix factorization and a residual graph attention network for denoising and embedding.

Main Results:

• FRGATMF significantly outperformed seven existing models on acupuncture and public datasets.
• The model demonstrated superior performance across various evaluation metrics.
• Identified previously unconsidered acupoint combinations with potential therapeutic benefits.

Conclusions:

• The FRGATMF model offers a robust and accurate method for predicting acupoint compatibility.
• This approach can expand treatment options by uncovering novel acupoint combinations.
• The findings highlight the potential of advanced computational methods in traditional medicine.