Dissertation

Explore the Relationship Between Genetic Variations and Phenotypes with Bayesian Approaches

Zhan, Jianan

2018

Overview

Genome-wide association studies (GWAS) have had great success in identifying human genetic variants associated with human traits. With recent developments in high throughput biology, immense amount of data have been generated, thus calling for novel statistical and computational approaches to be developed and draw biological meaningful conclusions. A current direction for GWAS method development has been to use Bayesian approaches, where prior beliefs of variant effects are incorporated into test statistics, to boost the power to detect real associations. With previous success in developing Bayesian-based GWAS method for single phenotype, in this work the Bayesian idea is extended to multiple phenotypes, aiming at developing a method that detects pleiotropic genome-wide associations. Alongside with the method development, analytical simulations were also performed to investigate into the possible power gain by using such Bayesian approaches, as well as to understand how different factors influence the behavior of Bayesian-based GWAS methods. Many variants are pleiotropic, and discovery of these variants could help reveal disease mechanisms, suggest new therapeutic options. Therefore, we developed a pleiotropic GWAS method based on Bayesian framework, SNP And Pleiotropic PHenotype Organization (SAPPHO), which learns pleiotropy using identified associations to discover additional associations with shared patterns. SAPPHO was applied on two sets of real data: 1. Atherosclerosis Risk in Communities (ARIC) study of 8,000 individuals, whose gold-standard associations were provided by meta-analysis of 40,000 to 100,000 individuals from the Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE) consortium; 2. Cancer phenotypes from UK Biobank project, consisting several hundred to 15,000 individuals, with gold-standard obtained from GWAS catalog. For both data sets, SAPPHO was able to detect additional associations that were not detected with the conventional univariate test, and boost power when different variants follow the same association patterns. Bayesian approaches boost power for GWAS through alleviating burdens from multiple hypothesis testings, which is usually on the scale of thousands to millions. Intuitively, by making use of prior probabilities that bias favored sets thought to be enriched for significant findings, power for detecting true associations could be increased. Therefore, an analytical study was conducted here to see theoretically to what extent power could gain by using such approaches, and how does this gain depend on different factors. By calculating test power assuming perfect knowledge of a prior distribution, the population size increase required to provided the same boost without a prior was obtained, and it is shown that population size is exponentially more important than prior, providing a rigorous proof for the lack of use for prior-based GWAS methods.

• Publisher: ProQuest Dissertations & Theses
• Subject: Biostatistics / Computer science / Genetics
• ISBN: 1392068827, 9781392068823