Causal mediation analysis for molecular mediators

There is growing interest in understanding the causal pathways linking exposures to health outcomes, with the goal of identifying potentially modifiable intermediate variables for intervention. This is done using mediation analysis - the study of chains of intermediate processes that link a risk factor to an outcome. A key priority in current research is understanding the degree to which high-dimensional molecular data (e.g. metabolomics, epigenetics) mediate the relationship between exposures and health outcomes.
One method to improve causal analysis within this framework is Mendelian Randomization (MR), which uses genetic variants as instrumental variables (IVs) to facilitate causal inference with observational data avoiding bias due to confounding. It has recently been shown that when genetic IVs are available for both a risk factor and a mediator, ‘network’ or ‘two-step’ MR can be used to estimate direct and indirect effects that are not affected by confounding.
This project will develop methods for use of MR for mediation studies with molecular intermediates, exploring questions such as: i) pleiotropy of instrumental variables for closely related mediators, ii) the use of instruments for composite/ratio variables, iii) the influence of measurement error in the mediator variable, iv) methods for assessing the joint and individual contributions of multiple related mediators. The methods developed will be applied to important clinical questions.
This PhD offers skills development in: big data, genetic and epigenetic epidemiology, causal analysis, statistical methods development. The student will be based in the MRC Integrative Epidemiology Unit, a world-leading centre for the development of causal analysis methods, with co-supervision from Dr Rhian Daniel (Cardiff).