Bayesian community detection for microbial interaction networks

The human body harbours diverse microbiomes that play a vital role in health, with microbial interactions providing key insights into many disease mechanisms. Uncovering the complex microbiome community structure is essential to understanding the role of the microbiome in disease progression and susceptibility. This paper introduces a novel weighted stochastic infinite block model for analyzing fully connected microbial interaction networks. Prior to model fitting, microbial interaction networks are constructed using a recently developed semi-parametric rank-based correlation method that accounts for zero-inflation and latent confounding during network construction. By adopting a Bayesian nonparametric approach, our model clusters taxa into distinct communities and estimates the number of communities without the need for ad hoc selection or post-processing. Through simulations and real data from postmenopausal women with recurrent urinary tract infections, we show our model achieves superior clustering accuracy over alternatives and uncovers several novel biological insights. Notably, this is the first study to explore urinary microbiome interaction networks while preserving all information from taxonomic abundance data.