A New Regression Model for the Analysis of Microbiome Data

Human microbiome data are becoming extremely common in biomedical research due to the relevant connections with different types of diseases. A widespread discrete distribution to analyze this kind of data is the Dirichletmultinomial. Despite its popularity, this distribution often fails in modeling microbiome data due to the strict parameterization imposed on its covariance matrix. The aim of this work is to propose a new distribution for analyzing microbiome data and to define a regression model based on it. The new distribution can be expressed as a structured finite mixture model with Dirichlet-multinomial components. We illustrate how this mixture structure can improve a microbiome data analysis to cluster patients into “enterotypes”, which are a classification based on the bacteriological composition of gut microbiota. The comparison between the two models is performed through an application to a real gut microbiome dataset.