Title: Sample size guided strategies for analysis of human microbiome data
Authors: Alexander Alekseyenko - Medical University of South Carolina (United States) [presenting]
Abstract: The human microbiome datasets capture the abundances and compositions of entire microbial communities inhabiting anatomical microenvironments, obtained under a variety of study designs and sample size constraints. Pairwise distance matrices, known as beta diversity measures, are often the basis for many analyses in the microbiome space. We will introduce descriptive multivariate methods based on visualization of beta diversities. Significance of observed patters is often determined with PERMANOVA. We will present new versions of the PERMANOVA test, Tw2 and Wd*, which overcome its adverse behavior under heteroscedasticity and sample size imbalance. Likewise, we will present PERMANOVA-S method, which allows to draw inferences from microbiome data using ensembles of beta-diversity indices. The above techniques are useful when the sample size is small to medium, but have reduced relative utility with increasing sample size. When the sample size allows for robust inferences about individual microbes and sub-communities of microbes a different set of techniques allow for better inference. We will review approaches with medium to large sample sizes based on univariate testing with methods that link association with predictivity and causality. Results on the performance of feature selection and machine learning methods with microbiome data will be presented.