Abstract:
|
The aim of this project is to identify CpG regulators that causally affect phenotype through regulating effect on gene expression (GE), and to assess what proportion of the total effect is mediated by GE, based on summary statistics. To achieve this, we can apply the principle of Multivariable Mendelian Randomization (MVMR) to GE as an intermediate exposure. However, reliability of MVMR depends on validity of the genetic variants, usually SNPs, as instrumental variables (IVs). SNPs that exhibit horizontal pleiotropy, if included in the MVMR, may bias the causal effect estimates. Therefore, we propose extending the Mendelian Randomization Pleiotropy RESidual Sum and Outlier (MR-PRESSO) test to a multivariable setting to detect and correct for pleiotropic outliers in MVMR. In our simulation study, the proposed method shows competitive power compared to the alternative Cook’s distance, Studentized residuals, and Q-statistics methods, especially when there is limited number of IVs. We further validate our method by applying to Framingham Heart Study summary statistics. We identify CpG exposures that causally affect pulmonary function through GEs and estimate the mediation proportions.
|