Statistical tools for the multi-omics analysis of microbiome data

Thesis (Ph.D.)--University of Washington, 2022The human microbiome consists of trillions of bacteria, archaea, and viruses that exist on virtually every organ in the body. The microbiome plays a fundamental role in human health and has been implicated in several different diseases and conditions such as cardiovascular disease and certain cancers. Understanding the functional role of the microbiome can lead to increased understanding of these complex diseases and result in the development of more effective treatments. Although advances in technology have allowed for the inexpensive processing and analysis of high-throughput data, several statistical challenges exist in the analysis of microbiome data. In my dissertation, I will present three projects that address the statistical challenges of high-dimensionality, multi-omics data integration, batch effects/other covariate adjustment, and the visualization of microbiome data. In Project 1, We address the issues of high-dimensionality and data integration by proposing a new procedure for testing the cumulative metabolic effect of the microbiome using a weighted variance component test framework. In this setup, we focus on metabolic pathways and recognize that metabolism can be represented by metagenomics (metabolic potential) and metabolomics (metabolic output). In Project 2, we address the issue of batch effects and high-dimensionality by outlining a two-step adjustment of the principal coordinates (PCs) of the microbial taxa data. In the first step, we project the mean effect of the unwanted covariates out of the PCs. In the second step, we adjust out the second moment of the same covariates from the PCs by assuming a linear relationship between the covariates and the variance of the PCs. Finally, in Project 3, we propose an effect modification testing procedure for evaluating interactions between microbial taxa and environmental factors on an outcome of interest. We address concerns of data integration and high-dimensionality by using a variance component test framework with LASSO-selected variables to assess the effect modification of the microbiome on environmental variables