Predicting disease-associated microbes based on similarity fusion and deep learning

Area of Science:

• Bioinformatics and Computational Biology
• Microbiology
• Genomics

Background:

• The human microbiome plays a critical role in numerous disorders.
• Identifying disease-associated microbes is crucial for understanding pathogenesis and treatment.
• Computational prediction of microbe-disease associations is a growing area of bioinformatics research.

Purpose of the Study:

• To present SGJMDA, a novel deep learning computational approach for predicting microbe-disease associations.
• To leverage advanced network analysis and feature extraction techniques for improved prediction accuracy.

Main Methods:

• SGJMDA fuses microbe and disease similarities using a nonlinear strategy.
• Feature extraction is performed using graph convolution networks on homogeneous networks.
• A heterogeneous microbe-disease network is constructed using multi-neighborhood graph convolution and jumping knowledge networks.

Main Results:

• SGJMDA outperforms six state-of-the-art computational methods in cross-validation experiments.
• Case studies on three diseases showed high accuracy, with 19, 20, and 11 top-20 predictions validated.
• The method demonstrates significant potential for inferring microbe-disease relationships.

Conclusions:

• SGJMDA is a valuable and promising tool for predicting microbe-disease associations.
• The approach enhances understanding of the microbiome's role in disease.
• This computational tool can guide future biomedical screening and therapeutic development.