Classifying mechanisms of pleiotropy to improve causal modelling

A decade of genome wide association studies (GWAS) has resulted in thousands of genetic variants known to associate with traits across the human phenome. With this resource comes the hope of a profound improvement in the drug discovery process, specifically by increasing the safety of new drugs and lowering their development costs. Mendelian randomisation (MR) is an experimental design that uses genetic associations with a putative 'exposure' phenotype to infer its causal effect on an 'outcome' phenotype of interest. Using known genetic associations in this fashion has proven to be effective at predicting the chance that a drug target (exposure) is likely to be effective for an outcome. It is now possible to construct causal networks of thousands to tens of thousands of molecular phenotypes that might aid in prioritising those molecular targets most likely to be causally influencing human physiology. One limitation of MR is its dependence on certain assumptions about genetic pleiotropy - it assumes that the influence of a genetic variant on two traits arises because it influences a primary exposure which in turn influences the outcome, but it is possible that pleiotropy can arise because the genetic variant influences the two traits through independent pathways. It is therefore essential to evaluate the extent to which this assumption does not hold, because this information could be integrated with MR methodology to create models that more reliably make causal inference. The focus of this PhD project will be to use a repository of billions of genetic associations across thousands of traits (MR-Base) to infer relative likelihoods of the different modes of pleiotropic actions.