Utilizing Dual-Channel Graph and Hypergraph Convolution Network to Discover Microbes Underlying Disease Traits.

Area of Science:

• Microbiology
• Computational Biology
• Bioinformatics
• Genomics

Background:

• Identifying microbes associated with diseases is crucial for developing new diagnostic and therapeutic strategies.
• Traditional experimental methods for discovering microbe-disease associations are time-consuming and expensive.
• There is a need for efficient computational approaches to accelerate the identification of these associations.

Purpose of the Study:

• To propose an innovative computational framework, the dual-channel graph and Hypergraph Convolutional Network (DCGHCN), for discovering microbes underlying disease traits.
• To address the limitations of traditional experimental methods by providing a faster and more cost-effective computational solution.
• To accurately predict microbe-disease associations (MDAs) and validate the model's effectiveness.

Main Methods:

• Constructed attribute graphs for microbes and diseases using the K-Nearest Neighbors (KNN) principle.
• Employed Graph Convolutional Networks (GCNs) to capture implicit representations from attribute graphs.
• Utilized a dual-channel structure combining GCNs and Hypergraph Convolutional Networks (HGCNs), incorporating KNN for missing value imputation.

Main Results:

• The DCGHCN model achieved high performance metrics: AUC of 0.9415, AUPR of 0.7637, F1-score of 0.7515, and accuracy of 0.9818, evaluated using 5-fold cross-validation.
• Case studies on two selected diseases demonstrated that the predicted microbe-disease associations were largely confirmed by existing literature.
• The model effectively handled missing values in the correlation matrix, showcasing its robustness.

Conclusions:

• DCGHCN is an effective and efficient computational tool for discovering microbes associated with disease traits.
• The dual-channel architecture integrating GCNs and HGCNs significantly enhances prediction accuracy.
• This approach offers a promising avenue for accelerating research in microbiome-based disease diagnosis and treatment.