Statistical approaches for causal analysis in genetics data

Mendelian randomisation (MR) analysis uses the random allocation of alleles in the population to obtain an "unconfounded" estimate of the association between a risk factor and an outcome. Increasingly, genetic studies of high dimensional data, such as proteomics, metabolomics or other "omics", is allowing wide-scale causal analysis using MR. However, the use of such data raises challenges for traditional methods; for example, many variants are pleiotropic and may therefore not make suitable tools for MR. Many methods have been proposed to disentangle these effects, such as proxy mediator approaches, and I will apply them to the rich genotypic and phenotypic data available in the department, such as the INTERVAL study, to provide more robust causal inference. The availability of a wide range of "omics" data will allow me to construct biologically informed subsets of genetic variants to use as instruments for MR for more interpretable results compared with agnostic approaches. Furthermore, I will augment the existing set of known associated variants with novel findings by undertaking genome wide association analysis leading to greater power for discovery of potential causal effects. I will develop and test other analysis strategies such as clustering based methods to help identify potential pathways across one or more genes to advance the understanding of specific biological processes that underpin the observed data structure and elucidate potential therapeutic drug targets.