STATISTICAL METHODS FOR GENETIC AND EPIGENETIC ASSOCIATION STUDIES Public Deposited

Downloadable Content

Download PDF
Last Modified
  • March 22, 2019
Creator
  • Huang, Kuan-Chieh
    • Affiliation: Gillings School of Global Public Health, Department of Biostatistics
Abstract
  • First, in genome-wide association studies, few methods have been developed for rare variants which are one of the natural places to explain some missing heritability left over from common variants. Therefore, we propose EM-LRT that incorporates imputation uncertainty for downstream association analysis, with improved power and/or computational efficiency. We consider two scenarios: I) when posterior probabilities of all possible genotypes are estimated; and II) when only the one-dimensional summary statistic, imputed dosage, is available. Our methods show enhanced statistical power over existing methods and are computationally more efficient than the best existing method for association analysis of variants with low frequency or imputation quality. Second, although genome-wide association studies have identified a large number of loci associated with complex traits, a substantial proportion of the heritability remains unexplained. Thanks to advanced technology, we may now conduct large-scale epigenome-wide association studies. DNA methylation is of particular interest because it is highly dynamic and has been shown to be associated with many complex human traits, including immune dysfunctions, cardiovascular diseases, multiple cancer, and aging. We propose FunMethyl, a penalized functional regression framework to perform association testing between multiple DNA methylation sites in a region and a quantitative outcome. Our results from both real data based simulations and real data clearly show that FunMethyl outperforms single-site analysis across a wide spectrum of realistic scenarios. Finally, large studies may have a mixture of old and new arrays, or a mixture of old and new technologies, on the large number of samples they investigate. These different arrays or technologies usually measure different sets of methylation sites, making data analysis challenging. We propose a method to predict site-specific DNA methylation level from one array to another - a penalized functional regression model that uses functional predictors to capture non-local correlation from non-neighboring sites and covariates to capture local correlation. Application to real data shows promising results: the proposed model can predict methylation level at sites on a new array reasonably well from those on an old array.
Date of publication
Keyword
Subject
DOI
Identifier
Resource type
Rights statement
  • In Copyright
Advisor
  • Chen, Mengjie
  • Sun, Wei
  • North, Kari
  • Lange, Ethan
  • Li, Yun
Degree
  • Doctor of Philosophy
Degree granting institution
  • University of North Carolina at Chapel Hill Graduate School
Graduation year
  • 2015
Language
Publisher
Place of publication
  • Chapel Hill, NC
Access
  • There are no restrictions to this item.
Parents:

This work has no parents.

Items