Identifying causal effects sizes in population genetics via Mendelian randomisation

Researchers, in the field of bio-medicine, are often interested identifying which set of biological exposures are causal of trait or disease.

Due to advances in the field of causal inference in recent years, both in terms of causal identifiability and estimation, the causal effect of a variable on an outcome in the presence of unobserved confounders can be estimated, under certain assumptions. A common technique to achieve this is known as the instrumental variable (IV) approach.

Here an additional variable is introduced and the exposure of interest is conditioned on this variable.

In the case where the instrument Z is a DNA variant, inherited at birth and essentially unchanging over a life time, the IV approach is known as Mendelian Randomisation (MR), a common approach to infer causality in population genetics.

Despite widespread application in the biomedical literature, certain assumptions underlying the validity of MR are often violated in practice.

One such violation is the DNA variant being used affects the outcome through some other pathway than just that of the exposure. This situation is termed, pleiotropy.

My project looks to apply Mendelian randomisation techniques to a range of exposure and outcome situations, as well as to develop statistically sound methodology for tackling the situation of pleiotropy - acknowledging that it is commonly accepted that one cannot easily identify which DNA variants may be pleiotropic.