|
Abstract:
|
Longitudinal studies of the microbiome are increasingly common as a way to separate within- and between-subject variability. They can also help clarify the association of health outcomes with either systematic shifts in composition or temporal instability in the microbiome (volatility). However, there is no clear framework for quantifying the extent of volatility in the microbiome to enable association testing with outcomes of interest. We propose formal measures of qualitative (appearance or disappearance) and quantitative (abundance) volatility, explore patterns of volatility in several real microbiome time series datasets, propose a volatility hypothesis testing framework, and perform extensive simulation studies to assess the performance of this framework under balanced and unbalanced designs, with dense or sparse time series/longitudinal datasets, and with or without confounding. We apply the approach to a real microbiome dataset in irritable bowel syndrome.
|
