MHGNN: Multiplex Hypergraph Neural Networks for Predicting Herb-Symptom Interactions.

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Herb-symptom interaction (HSI) prediction is vital for understanding herbal therapies and advancing precision traditional Chinese medicine (TCM).
• Current graph neural network (GNN) methods for HSI prediction are limited to pairwise associations and struggle with high-order effects.
• Incomplete data leads to false negatives in random negative sampling for existing models.

Purpose of the Study:

• To develop a novel hypergraph learning framework, MHGNN, for more accurate HSI prediction.
• To address limitations of existing GNNs in capturing high-order herb-symptom relationships and mitigate false-negative bias.
• To enhance mechanism-aware modeling for precision herbal medicine.

Main Methods:

• MHGNN models HSIs using a multiplex hypergraph integrating protein-protein interactions (PPIs) and high-order herb-protein/symptom-protein relationships.
• Hierarchical message passing aggregates high-order features across multiple edge types.
• A network-based negative sampling strategy selects distant herb-symptom pairs to reduce false negatives.

Main Results:

• MHGNN demonstrated superior predictive performance compared to thirteen state-of-the-art baseline methods on two public TCM datasets.
• The model effectively captures high-order and indirect regulatory effects in herb-symptom relationships.
• The network-based negative sampling strategy successfully reduced false-negative bias.

Conclusions:

• MHGNN offers a significant advancement in HSI prediction by leveraging hypergraph learning.
• The framework enhances mechanism-aware modeling, paving the way for data-driven precision herbal medicine.
• MHGNN's ability to model complex biological networks holds potential for future pharmacological research.