MNetClass: A Control-Free Microbial Network Clustering Framework for Identifying Central Subcommunities Across Ecological Niches

Investigating microbiome subnetworks and identifying central microbes in specific ecological niches is a critical issue in human microbiome studies. Traditional methods typically require control samples, limiting the ability to study microbiomes at distinct body sites without matched controls. Moreover, some clustering methods are not well-suited for microbial data and fail to identify central subcommunities across ecological niches after clustering.

In this study, we present MNetClass, a novel microbial network clustering analysis framework. It utilizes a random walk algorithm and a rank-sum ratio-entropy weight evaluation model to classify key subnetworks and identify central microbes at any body site without the need for control samples. By applying MNetClass to microbiome data from five distinct oral sites, we successfully uncovered site-specific microbial subgroups and their central microbes. Additionally, simulations and Autism Spectrum Disorder (ASD) cohort datasets demonstrated that MNetClass outperforms existing unsupervised microbial clustering methods in terms of both accuracy and predictive power. In case studies, MNetClass identified age-related microbial communities across different oral sites, highlighting its broad applicability in microbiome research.