Statistical Methods for the Estimation of Cell-Type Composition and Cell-Type Specific Association Studies Public Deposited

Downloadable Content

Download PDF
Last Modified
  • March 20, 2019
Creator
  • Wilson, Douglas
    • Affiliation: Gillings School of Global Public Health, Department of Biostatistics
Abstract
  • Samples of human tissues used in biological research are often impure. Such samples contain cells of the type under study and multiple ancillary cell types, leading to inaccurate expression estimates and analysis for the cell type under study. While estimates of cell type abundance can be of interest on their own, their use is critical to the correction of differential expression testing in heterogeneous cell type samples to account for differential cell type abundance across conditions. This dissertation develops and examines three statistical models for the estimation or use of cell type abundance profiles in the analysis of RNA-seq data from heterogeneous cell type samples. Regarding estimation of cell type abundance profiles, we propose two models: IsoDeconv and ICeD-T. The IsoDeconv model approaches abundance estimation using isoform-level expression. We extend the IsoDeconv model to allow for biological variability in isoform expression across samples. The IsoDeconv model is assessed via simulation and through use of in silico mixtures of genuine RNA-seq expression datasets from non-cancerous human cell lines. The ICeD-T model approaches abundance estimation deconvolution using gene-level expressions while allowing for aberrant gene behavior within mixed cell type samples. Estimation properties of ICeD-T are assessed via simulation and validated in both microarray and RNA-seq datasets. Transitioning to the use of abundance profiles in the analysis of heterogeneous cell type samples, we propose pTReCASE. pTReCASE is an expression quantitative trait locus (eQTL) mapping technique for use in bulk tumor samples. pTReCASE extends current eQTL mapping methods for tumor tissues to estimate eQTLs within tumor and normal cells separately. The type I error and power of pTReCASE are assessed via simulation before application to the study of breast cancer data from 547 Caucasian women.
Date of publication
Keyword
DOI
Resource type
Rights statement
  • In Copyright
Advisor
  • Sun, Wei
  • Ibrahim, Joseph
  • Wu, Di
  • Li, Quefeng
  • Rashid, Naim
Degree
  • Doctor of Philosophy
Degree granting institution
  • University of North Carolina at Chapel Hill Graduate School
Graduation year
  • 2018
Language
Parents:

This work has no parents.

Items