All ETDs from UAB

Advisory Committee Chair

Nengjun Yi

Advisory Committee Members

Charles Katholi

Leann Long

Peter Mannon

Casey Morrow

Document Type

Dissertation

Date of Award

2019

Degree Name by School

Doctor of Philosophy (PhD) School of Public Health

Abstract

The human microbiome was first described as an “ecological community of commensal, symbiotic, and pathogenic microorganisms that literally share our bodyspace”. These microorganisms, also known as microbes and bacteria, are known for participation in human health. For example, microbes extract energy from foods that are otherwise indigestible, synthesize vitamins and amino acids necessary for important bodily functions, and protect hosts against disease-carrying pathogens. Therefore, the pursuit of human microbiome research is important for understanding the influence of microorganisms throughout human systems. Human microbiome research is founded on advances in next-generation sequencing technology. Namely, the identification of microbes that constitute the microbiota and associated metagenome drives interpretations of interactions and variability within and between microbes residing on and in human subjects. Relationships between bacteria and disease have been identified, for example, in cases of inflammatory bowel disease, diabetes, periodontal disease and various cancers. Also, evidence has been found to suggest changes associated with demographic traits such as age and race and behavioral traits such as dietary practices and antibiotic usage. A number of important questions concerning the commonalities and divergences in the human microbiome, however, have yet to be addressed. The primary interest of this dissertation involves addressing these outstanding questions through differential abundance analysis which is utilized to determine whether microbes are statistically associated with host characteristics. The simplicity of this goal is complicated due to characteristics of human microbiome count data such as fluctuating library size, over-dispersion, high-dimensionality, and sparsity. Fortunately, the first challenges have been widely studied in the context of microarray and RNA-Seq experiments. In order to address high-dimensionality, we first propose a Bayesian hierarchical negative binomial model capable of providing a comprehensive solution in respect to regression coefficient estimation when the number of covariates equals or exceeds the number of samples. Moreover, we propose a Bayesian hierarchical zero-inflated negative binomial model to provide even better fit by correcting for sparsity through delineation between sampling and true zeros. Application to the American Gut Project demonstrates the usefulness of the proposed methodologies given the ability to identify biologically relevant microbes consistent with contemporary literature.

Included in

Public Health Commons

Share

COinS