Simple Bayesian approaches to modelling pleiotropy in genetic association data

Thesis

Genome-wide association studies (GWAS) have become one of the most common types of genetic studies, due to their success at finding markers associated with diseases and other traits of interest. With thousands of these studies performed in the last decade, a variety of methods have been developed to meta-analyze their results to search for variants that affect multiple traits. This thesis introduces one such method, based on approximate Bayes factors (ABFs), which relies only on effect size estimates and standard errors from GWAS. Through application to simulated data as well as three different datasets, we demonstrate the statistical properties, strengths, and limitations of our approach.

We show that only can this method be applied to the meta-analysis of a single trait across multiple studies, but because it does does not make strong assumptions about the similarity of effect sizes across traits, it can also be used to detect effects across multiple traits at a single marker. Additionally, it can account for confounding due to things like shared samples and can be applied exhaustively across all possible combinations of associations to determine the subset of traits or studies that are most likely to be associated with a given variant. This affords the opportunity to make statements about which traits explicitly are and are not associated with a marker, and these patterns can be explored over the whole genome to learn about the genetic relationships between different traits. We also discuss some of the individual markers highlighted by our analyses—some known, and some potentially novel—and the traits associated with them.

Authors: Trochet, H

• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford