Date of Award

2019

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Public Health Sciences

College

College of Graduate Studies

First Advisor

Elizabeth Hill

Second Advisor

Dongjun Chung

Third Advisor

Richard Drake

Fourth Advisor

Mulugeta Gebregziabher

Fifth Advisor

Andrew Lawson

Sixth Advisor

Elizabeth Yeh

Abstract

Matrix-assisted laser desorption/ionization Fourier-transform ion-cyclotron resonance (MALDI FT-ICR) imaging mass spectrometry (IMS) technology allows researchers to measure the abundance of ionized fragments over a two-dimensional space. Despite advances in IMS technology, methods used to analyze such data have lagged. In particular, the variability in IMS data can be attributed to both spatial and random sources. Additionally, the frequency of masses with high proportions of zero abundance measures is often quite large. To address these issues, we automate a procedure to account for spatial variability across multiple regions of interest. Using that procedure, we then develop and propose log-linear regression models facilitating group-level comparisons of ionized fragment abundance, which further account for both the data's spatial structure and excess zeros. Our regression models, while accounting for the spatial structure in the same way, differ in their assumptions about the nature of the zeros, in particular whether they are accurately measured zeros or left-censored observations. We evaluate our models using simulated data and compare performance to approaches that account for the spatial information with differing complexity. We demonstrate that our methods maintain lower type I error rates and higher coverage compared with other approaches. These trends become more pronounced with increasing proportions of zeros, whether those zeros be true zero abundance measures or censored observations. We apply our models to a study examining glycosylation patterns in metastatic breast cancer. We identify N-glycans with differential abundance between primary and secondary tumor tissues, as well tissues stained negative and positive for tumor-associated macrophages. Upon classifying N-glycans into functional groups, we identify patterns that suggest underlying changes in enzymatic activity. Lastly, we develop the R package imagingPC that utilizes our methods to make them accessible to investigators. The R package distills our methods into a small set of functions that require limited knowledge of R. In addition to the base functions that use our methods, we incorporate functions that make our approach transparent and allow users to assess model assumptions.

Rights

All rights reserved. Copyright is held by the author.

Download

Share

COinS