RARE GENETIC VARIANTS AND COMPLEX DISEASES - A BAYESIAN APPROACH

Show full item record

Permalink

http://urn.fi/URN:ISBN:978-951-51-1671-0
Title: RARE GENETIC VARIANTS AND COMPLEX DISEASES - A BAYESIAN APPROACH
Author: He, Liang
Contributor: University of Helsinki, Faculty of Medicine, Hjelt Institute
Publisher: Helsingin yliopisto
Date: 2015-11-06
URI: http://urn.fi/URN:ISBN:978-951-51-1671-0
http://hdl.handle.net/10138/157150
Thesis level: Doctoral dissertation (article-based)
Abstract: Statistical inference of genome-wide association studies (GWAS) on a variety of epidemiological phenotypes with special focus on rare variants (RVs) is challenging. In this thesis, new statistical models for detecting RV association have been developed under the hierarchical Bayesian framework. Special attention with respect to appropriate statistical inference has been given to the case of incorporation of sequencing error information, family-based models and detection of interaction effects in twin data. Estimation of the three new statistical models proposed in this thesis have been implemented using Markov chain Monte Carlo (MCMC) methods and their statistical properties have been evaluated through simulations. Their performances have been compared with other generally used statistical methods. In effect, through hierarchical Bayesian modeling that incorporates more complex settings, some of these models are shown to be superior to the other methods in certain scenarios in terms of statistical power and robustness in identifying RV association. All models have also been applied to real data analyses to detect RVs significantly associated with continuous phenotypes such as systolic blood pressure and low-density lipoprotein cholesterol level. Some of our results confirm previous findings and others provide novel evidence of the involvement of RVs in these complex phenotypes that are missed by other methods such as SKAT and Granvil. Additionally, we focus on applying a time-to-event model with a kinship matrix to GWAS on transitions between different smoking stages to improve our understanding of the genetic architecture underlying smoking behavior. We employ a Cox model with multivariate normal random effects to deal with correlated time-to-event phenotypes. Our results provide novel evidence for supporting the hypothesis that complex neurotransmitter networks are involved in initiation of smoking behavior.Statistical inference of genome-wide association studies (GWAS) on a variety of epidemiological phenotypes with special focus on rare variants (RVs) is challenging. In this thesis, new statistical models for detecting RV association have been developed under the hierarchical Bayesian framework. Special attention with respect to appropriate statistical inference has been given to the case of incorporation of sequencing error information, family-based models and detection of interaction effects in twin data. Estimation of the three new statistical models proposed in this thesis have been implemented using Markov chain Monte Carlo (MCMC) methods and their statistical properties have been evaluated through simulations. Their performances have been compared with other generally used statistical methods. In effect, through hierarchical Bayesian modeling that incorporates more complex settings, some of these models are shown to be superior to the other methods in certain scenarios in terms of statistical power and robustness in identifying RV association. All models have also been applied to real data analyses to detect RVs significantly associated with continuous phenotypes such as systolic blood pressure and low-density lipoprotein cholesterol level. Some of our results confirm previous findings and others provide novel evidence of the involvement of RVs in these complex phenotypes that are missed by other methods such as SKAT and Granvil. Additionally, we focus on applying a time-to-event model with a kinship matrix to GWAS on transitions between different smoking stages to improve our understanding of the genetic architecture underlying smoking behavior. We employ a Cox model with multivariate normal random effects to deal with correlated time-to-event phenotypes. Our results provide novel evidence for supporting the hypothesis that complex neurotransmitter networks are involved in initiation of smoking behavior.
Subject: public Health, Biostatistics
Rights: This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.


Files in this item

Files Size Format View

There are no files associated with this item.

This item appears in the following Collection(s)

Show full item record