Bayesian structural learning of microbiota systems from count metagenomic data
share
Metagenomics combined with high-resolution sequencing techniques have enabled researchers to study the genomes of entire microbial communities. Unraveling interactions between these communities is of vital importance to understand how microbes influence human health and disease. However, learning these interactions from microbiome data is challenging, due to the high dimensionality, discreteness, broad dispersion levels, compositionality and excess of zero counts that characterize these data. In this paper, we develop a copula graphical model for structure learning in these settings. In particular, we advocate the use of discrete Weibull regression for linking the marginal distributions to external covariates, which are often available in genomic studies but rarely used for network inference, coupled with a Gaussian copula to model the joint distribution of the counts. An efficient Bayesian procedure for structural learning is implemented in the R package BDgraph and returns inference of the marginals and of the dependency structure, providing simultaneous differential analysis and graph uncertainty estimates. A simulation study and a real data analysis of microbiome data show the usefulness of the proposed approach at inferring networks from high-dimensional count data in general, and its relevance in the context of microbiota data analyses in particular.
READ FULL TEXT