The genetics of participation and some aspects of genetic data

Thesis

One of the major challenges in human genetic studies is to differentiate between results that reflect biological mechanisms and those that reflect data specific properties. If genetics play a role in determining who would participate in a study that includes DNA collection, the genotypes of the participants would not be representative of the target population. By its very nature, this bias cannot be investigated by comparing genotypes of participants and non-participants. In this thesis I introduce a new method that enables one to investigate participation bias using only the genetic samples of the participants. The method involves comparing shared and non-shared alleles of first-degree relatives and is sensitive only to direct genetic effects, i.e. there is no confounding with population stratification or indirect genetic effects. Applying this method to perform a genome-wide association analysis using data of first-degree relative pairs in the UK Biobank, I demonstrate simultaneously that the method works and there is a very substantial genetic component underling participation in UK Biobank. The demonstration is achieved by showing that a polygenic score constructed from the association analysis can, in a non-overlapping sample, significantly predict traits and behaviour that include educational attainment, BMI, age at first birth and participation in follow-up studies. The method includes three separate test statistics that each captures real participation effects, but are affected by data artefacts in a very different manner. Consequentially, data artefacts could potentially be identified through comparison of the three statistics, an alternative utility of the method. The thesis also includes investigating the effectiveness of using principal components to adjust for population stratification, and the many possible causes of deviation from Hardy-Weinberg equilibrium. Together with studying participation bias, these investigations have a common goal of potentially improving the many analyses that are, or can be, performed using a large dataset such as the UK Biobank.

Authors: Benonisdottir, S

• DOI: 10.5287/ora-yx8ryj6no
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: participation bias; genetic studies
• Subjects: Genetics--Statistical methods; Genetics; Participation