(MRC DTP) Identification of Causal Genes from Integrative Analysis of Association Studies using Colocalization and Mendelian Randomization

Cardiovascular disease (CVD) is the most common cause of death. However, the majority of CVD can be prevented by early interventions such as changing life-styles or taking medical treatments. It is believed that certain variables (e.g. genes) can cause CVD. To date, genome-wide association studies have identified a number of genetic variants that are predictive of CVD. It is crucial to further understand the underlying mechanisms as to which genes can potentially be causal and what genetic variables (mediators) convey such causal effect on CVD for the development of medical treatments.

Mendelian randomization approach has attracted much attention in recent years, with the primary aim of investigating causal relationships between mediators and diseases. In this approach, mediator associated variants are used as instruments which require three assumptions. However, two of these assumptions are untestable because of unmeasured confounders. Thus, a major challenge is identification of genetic variants that could possibly serve as valid instruments.

More recently, statistical colocalization methods have emerged in genetic analysis, aiming at examining causal pathways from genetic variants to mediators, and eventually, to diseases. This approach hypothesizes that if a mediator is on the causal pathway from genetic variants to disease, one would expect variants predictive of the mediator are also predictive of the disease. Statistical colocalization is tailored to search for common signals by integrating the results from mediator association and disease association analyses, to identify candidate causal pathways.

Despite the above differences, we see similarities between the two approaches. For example, ratio estimator is considered as an option in both Mendelian randomization and statistical colocalization [1,2].

This project shall systematically investigate the two approaches, from the specific causal questions addressed, underlying assumptions required to statistical methods applied. The student is expected to develop a statistical model to examine causality by combining the two methods in an appropriate way, which will then be applied to cardiovascular data of interest.