|
Abstract:
|
The analysis of Microbiome data is particularly challenging, due to specific characteristics of the data. For example, the type and frequency of units observed varies widely across subjects and, longitudinally, within a subject. In this work, we discuss a class of Bayesian Nonparametric mixture processes and discuss its application to problems in large-scale hypothesis testing. More specifically, we consider mixtures of Two Parameter Poisson-Dirichlet Processes (TPPDP), and describe their properties with specific reference to the random partition structures they induce. Furthermore, we describe how this type of mixtures can be conveniently used in multiple testing for Microbiome data to compare taxa distributions over multiple conditions. Theoretical results, computational details, and a comparison with recent competing methods are also provided.
|
