Detection of Horizontal Pleiotropy in Mediation Analysis of Omics Layers on Complex Traits
Conference
64th ISI World Statistics Congress - Ottawa, Canada
Format: IPS Abstract
Session: IPS 230 - Recent Developments in Statistical Genetics and Genomics
Thursday 20 July 2 p.m. - 3:40 p.m. (Canada/Eastern)
Abstract
High-dimensional omics datasets provide valuable resources to determine the causal role of molecular traits in mediating the path from genotype to phenotype. By integrating quantitative trait loci (QTL) and genome-wide association studies (GWAS) summary statistics, multivariable Mendelian randomization (MVMR) allows for mediation analysis to quantify the connectivity between omics layers and their causal impact on complex traits and diseases. This approach retains the benefits of using genetic instruments for causal inference, such as avoiding bias due to confounding, while allowing for the estimation of the different effects required for mediation analysis. However, there are several assumptions required for MVMR to give reliable estimates of the mediation proportion, including the absence of horizontal pleiotropy, which occurs when the variant affects the outcome through mechanisms other than via its effect on the set of exposures. Violation of the “no horizontal pleiotropy” assumption can cause bias in MVMR. We propose a robust approach for MVMR that identifies and eliminates horizontal pleiotropy, resulting in unbiased mediation estimates and higher power to detect causal relationships. We evaluated our approach under various simulation settings and compared it to several other approaches to detect pleiotropy. Moreover, we validated our approach with an application to the Framingham Heart Study using DNA methylation data to reflect exposures that causally affect pulmonary function through gene expression.