MNetClass: a control-free microbial network clustering framework for identifying central subcommunities across...

Area of Science:

• Microbiome research
• Network analysis
• Computational biology

Background:

• Investigating microbiome subnetworks and central microbes is crucial for understanding human health.
• Traditional methods often require control samples and struggle with microbial data clustering.
• Existing methods may fail to identify central subcommunities across different ecological niches.

Purpose of the Study:

• To introduce MNetClass, a novel framework for microbial network clustering analysis.
• To identify central microbes and key subnetworks at any body site without control samples.
• To demonstrate the broad applicability and superior performance of MNetClass in microbiome research.

Main Methods:

• Utilizes a random walk algorithm for network analysis.
• Employs a rank-sum ratio-entropy weight evaluation model for classification.
• Applies MNetClass to simulated and real microbiome data from distinct body sites.

Main Results:

• MNetClass outperforms current unsupervised microbial clustering methods on simulated data.
• Analysis of oral microbiome data revealed site-specific microbial communities.
• Demonstrated superior predictive performance on Autism Spectrum Disorder data and identified age-related communities.

Conclusions:

• MNetClass is a valuable tool for microbiome network analysis, identifying key microbial subcommunities.
• The R package offers broad accessibility for researchers.
• MNetClass provides a robust framework for advancing microbiome research without control groups.