Dissertation/Thesis Abstract

Adaptive Statistical Methods for Microbiome Association Studies
by Koh, Hyunwook, Ph.D., New York University, 2018, 118; 10750033
Abstract (Summary)

The human microbiome studies have been accelerated by the advances in next-generation sequencing technologies. There has also been increasing interest in discovering microbial taxa that are associated with diverse host phenotypes, environmental factors or clinical interventions. Here, I first describe unique features of microbiome data and the resulting demand for adaptive association analysis which robustly suits different association patterns, while providing valid statistical inferences. Then, I introduce two adaptive microbiome association tests as follows.

My first method, namely, optimal microbiome-based association test (OMiAT), relates microbial composition with continuous (e.g., body mass index) or binary (e.g., disease status) traits. OMiAT is a data-driven adaptive testing method which approximates to the most powerful performance among different candidate tests from the sum of powered score tests (SPU) and microbiome regression-based kernel association test (MiRKAT). I illustrate that OMiAT robustly discovers underlying association signals arising from highly imbalanced microbial abundances and phylogenetic tree structure, while correctly controlling type I error rates. I also propose a way to apply it to fine association mapping of diverse higher-level taxa at different taxonomic levels within a newly introduced microbial taxa discovery framework, microbiome comprehensive association mapping (MiCAM).

My second method, namely, optimal microbiome-based survival analysis (OMiSA), relates microbial composition with survival (i.e., time to event) traits. OMiSA approximates to the most powerful association test within two test domains, 1) microbiome-based survival analysis using linear and non-linear bases of OTUs (MiSALN) and 2) microbiome-based kernel association test for survival traits (MiRKAT-S). I illustrate that OMiSA powerfully discovers underlying associated lineages whether they are rare or abundant and phylogenetically related or not, while correctly controlling type I error rates.

OMiAT and OMiSA are attractive in practice due to the high complexity of microbiome data and the unknown true nature of the state. MiCAM also provides a hierarchical microbiome association map through a breadth of taxonomic levels, which can be used as a guideline for further investigation on the roles of discovered taxa in human health or disease.

Indexing (document details)
Advisor: Li, Huilin
Commitee: Blaser, Martin J., Goldberg, Judith D., Shao, Yongzhao, Troxel, Andrea B.
School: New York University
Department: Environmental Health Science
School Location: United States -- New York
Source: DAI-B 79/12(E), Dissertation Abstracts International
Source Type: DISSERTATION
Subjects: Biostatistics
Keywords: Adaptive association analysis, Compositional data analysis, High-dimensional statistics, Microbial group analysis, Microbiome association studies, Phylogenetics
Publication Number: 10750033
ISBN: 9780438170919
Copyright © 2019 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy
ProQuest