Oxofrd, Genomic Medicine and Statistics. (360G-Wellcome-109106_Z_15_Z)

One goal in the post Genome Wide Association Study (GWAS) era lies in characterizing gene-environment (G-E) interactions. These interactions are typically defined as a joint effect of one or more genes with one or more environmental factors that cannot be explained by their marginal effects. It is thought that better characterization of these interactions could be utilized in several ways [1]. By allowing populations to be stratified into subsets based on their response to certain environments, it may allow more informed public health decisions. It may also provide an answer to the question of missing heritability [2]. More optimistically, it may lead to the discovery of novel genes that act in concert with either the environment or other genes to change disease risk, but have no demonstrable marginal effects themselves. To date there has some interest in testing for G-E interactions in animal models, for example looking at schizophrenia in mouse models [3-5], but there has been limited success in uncovering examples of G-E interactions in humans. This is partly due to the problem of exposure assessment [1]. The exposure intensity of many environmental factors (for example, the composition of harmful particles in air pollution) varies over time. Even if this is not the case, disease risk is still likely to be modified by temporal factors such as the age at exposure or duration of exposure. In short, outside of animal models it has historically proved difficult to obtain data that accurately summarizes the environment of a number of individuals over a lifetime. The release of data from the UK Biobank in 2016 changed this. The UK Biobank is a major collaborative research project with data on over 500,000 volunteers aged 40-69 from the UK. The project aims to provide a comprehensive and reliable quantification of the combined effects of lifestyle, environment, genes and other exposures on health outcomes [6]. The UK Biobank not only represents a 2-fold increase in the number of participants that have typically been involved in genomic studies, but it also contains one of the most comprehensive databases on environmental factors collected to date – including detailed work history, full body MRI, dietary questionnaire responses and medical history. This PhD will focus on developing scalable statistical methods to detect G-E interactions, using the UK BioBank as a model dataset. (b) Previous research done on this topic area within the Marchini Group includes an investigation into using Gaussian Process Regression (GPR) to estimate missing heritability [7]. This approach was tested on 46 yeast phenotypes, and found that over 70% of the total known missing heritability could be explained from common genetic variants. Using a technique known as Automatic Relevance Determination [8], this method can be adapted to act as a classifier to identify environmental & genetic covariates that are associated with the trait. Initial results suggest that this method is also able to detect non-linear associations. A rotation project implemented this approach in the statistical programming language [R] during the first year of the GMS programme. (c) Approach and Methods to be used This project will follow a Bayesian methodology in attempting to infer G-E interactions. We will first start by testing known models on the dataset, such as multivariate linear regression (in effect a null model) or a random effects model, before progressing on to fitting a model based on GPR. As the size of the dataset presents a significant computational challenge, the majority of this will be programmed in C++. (d) Contingency Plans We cannot guarantee that we will uncover novel G-E interactions in the UK Biobank dataset, although due to the size of the dataset we think the chance is high. If not, then these results will still be of great interest to the community. Simulation studies will also be used to assess the power of the methods under different conditions, which will inform study design. (e) Initial milestones for the project include: \- Developing expertise with the programming language C++ \- Gaining familiarity with the UK BioBank and collating a first test dataset. We have identified that the blood pressure phenotypes will be good candidates for testing due to their continuous nature, large number of known marginal genetic associations and likelihood of environment playing a role in the phenotypes. \- Gaining an understanding of mathematical derivation of possible methods. \- Extending knowledge of the literature. Beyond this, over the next twelve months we aim to have implemented several approaches, including a multivariate linear regression and a random effects model, on the UK BioBank in time for the transfer of status viva.