A Generalized Bayesian Stochastic Block Model for Microbiome Community Detection.

Area of Science:

• Microbiome research
• Statistical modeling
• Bioinformatics

Background:

• Next-generation sequencing accelerates microbiome studies, increasing quantitative network analysis.
• Understanding microbiome community structure is key to disease research.
• Metagenomic data present challenges: high-dimensionality, compositional nature, uneven depth, over-dispersion, and zero-inflation.

Purpose of the Study:

• To propose a novel statistical method for microbiome co-occurrence network analysis and community detection.
• To address the challenges of analyzing high-dimensional, compositional microbiome data.
• To leverage taxonomic information for improved microbiome community structure inference.

Main Methods:

• Developed a generalized Bayesian stochastic block model tailored for microbiome data.
• Applied modified centered-log ratio transformation to microbiome abundance data.
• Incorporated taxonomic tree information using a Markov random field prior.
• Utilized Markov chain Monte Carlo sampling for joint parameter inference.

Main Results:

• The proposed model outperforms competing methods in simulation studies, even without informative taxonomic tree data.
• Successfully applied the method to a real urinary microbiome dataset from postmenopausal women.
• Revealed the urinary microbiome co-occurrence network structure in postmenopausal women for the first time.

Conclusions:

• The generalized Bayesian stochastic block model offers a robust approach for microbiome community detection.
• This statistical methodology provides a valuable new tool for advanced microbiome studies.
• The findings open new avenues for understanding the role of the urinary microbiome in postmenopausal health.