Genetics and causality: towards more accessible and more reliable Mendelian randomization investigations (360G-Wellcome-204623_Z_16_Z)

I propose to advance methods for Mendelian randomization to make investigations more reliable and more accessible, and to build a team to develop and apply these state-of-the-art methods. I will continue with the development of robust methods that give consistent estimates even when the stringent instrumental variable assumptions are not fully satisfied, and compare how these methods perform with real data. I will extend existing methods, considering estimates of non-linear causal relationships, and approaches for variable selection with heterogeneous genetic variants in different gene regions, and with highly-correlated variants from the same gene region. I will develop novel approaches for using covariate matching and matching by design (such as the analysis of sibling pairs). I will disseminate these methods in explanatory papers aimed at applied researches, and in a software package. I will partner with leading epidemiological and clinical researchers to apply these methods to scientific questions of interest, and feedback difficulties from these analyses into further methods development. I will develop pipelines for the prioritization of biomarkers as targets for pharmaceutical or clinical intervention from high-dimensional datasets, where thousands of candidate risk factors have been measured and detailed analysis of each risk factor in turn is impractical.