Sometimes Correlation does Equal Causation: Developing Statistical Methods to Determine Causality Using Genetic Data

There is a well-known mantra that correlation does not necessarily equal causation. This is why randomized controlled trials in which participants are physically randomized into treatment and placebo groups are the gold standard for assessing causality in epidemiological investigations. However, what is less appreciated is that strong evidence for causality can sometimes be obtained using observational data only. In particular, genotypes are randomly transmitted from parents to their offspring independent of the environment and other confounding factors, meaning that genotypes associated with particular traits can be used like natural “randomized controlled trials” to examine whether these traits causally affect risk of disease.

The aim of this PhD project is to develop statistical methods to assess causality using observational data alone. The successful candidate will gain experience across a wide range of advanced statistical genetics methodologies including Mendelian randomization (a way of using genetic variants to investigate putatively causal relationships), structural equation modelling, genome-wide association analysis (GWAS), genetic restricted maximum likelihood (G-REML) analysis of genome-wide data which can be used to partition variation in phenotypes into genetic and environmental sources of variation, and instrumental variables analysis (using natural “experiments” to obtain information on causality from observational data). The candidate will apply the new statistical methods that they develop to large genetically informative datasets like the UK Biobank (500,000 individuals with genome-wide SNP data).

*Qualifies for an earmarked scholarship