|
Abstract:
|
Great strides have been made in studying longitudinal relationships between the microbiome (gut, throat, skin, etc.) and human health (cancer, inflammatory markers, etc.) both at the individual taxon level and community level. However, current longitudinal methods for analyzing beta diversity, the measure of compositional similarity between two microbiome samples, are limited. Methods that use the full diversity matrix rarely model temporal trends in diversity. Instead, methods designed for temporal trends in diversity focus on a subset of diversity values ( e.g. Consecutive timepoints within subjects) to identify trends of interest. Here we present an extension of kernel association tests to detect temporal trends in beta diversity while using the full beta diversity matrix to identify the utility of incorporating every diversity value. In real data and simulations modeling a variety of temporal trends, we compare the power and type I error of our method to existing beta diversity association tests (PERMANOVA, CSKAT, GLMMs). We identify scenarios in which methods are optimal enabling microbiome studies to choose the most powerful to address the research question(s) of interest.
|
