Predicting microbe-disease associations via graph neural network and contrastive learning.

Area of Science:

• Microbiology and Bioinformatics
• Computational Biology and Health Informatics

Background:

• Growing recognition of microbes' role in human health.
• Limitations of traditional experimental methods for microbe-disease association studies (time-consuming, costly).
• Need for efficient computational approaches to predict microbe-disease links.

Purpose of the Study:

• To develop a novel computational framework, GCATCMDA, for predicting potential microbe-disease associations.
• To overcome the limitations of traditional experimental validation methods.

Main Methods:

• Construction of Gaussian kernel similarity networks for microbes and diseases.
• Utilizing a feature encoder combining graph convolutional networks and graph attention mechanisms.
• Employing a feature dual-fusion module for integrating node features.
• Applying contrastive learning to enhance feature consistency across networks.
• Using an inner product decoder for association score calculation.

Main Results:

• GCATCMDA demonstrated superior predictive performance compared to existing methods.
• Experimental results validated the model's effectiveness.
• Case studies confirmed GCATCMDA's utility in real-world prediction scenarios.

Conclusions:

• GCATCMDA is an effective computational tool for predicting microbe-disease associations.
• The framework offers a promising approach for advancing research in microbial roles in health and disease.