Bayesian Inference of Microbial Community Structure from Metagenomic Data Using BioMiCo

Abstract

Microbial samples taken from an environment often represent mixtures of communities, where each community is composed of overlapping assemblages of species. Such data represent a serious analytical challenge, as the community structures will be present as complex mixtures, there will be very large numbers of component species, and the species abundance will often be sparse over samples. The structure and complexity of these samples will vary according to both biotic and abiotic factors, and classical methods of data analysis will have a limited value in this setting. A novel Bayesian modeling framework, called BioMiCo, was developed to meet this challenge. BioMiCo takes abundance data derived from environmental DNA, and models each sample by a two-level mixture, where environmental OTUs contribute community structures, and those structures are related to the known biotic and abiotic features of each sample. The model is constrained by Dirichlet priors, which induces compact structures, minimizes variance, and maximizes model interpretability. BioMiCo is trained on a portion of the data, and once trained a BioMiCo model can be employed to make predictions about the features of new samples. This chapter provides a set of protocols that illustrate the application of BioMiCo to real inference problems. Each protocol is designed around the analysis of a real dataset, which was carefully chosen to illustrate specific aspects of real data analysis. With these protocols, users of BioMiCo will be able to undertake basic research into the properties of complex microbial systems, as well as develop predictive models for applied microbiomics.