Massively parallel multi-mediator analyses with integrative genomic application

Abstract

Mediation analysis is performed to evaluate the effects of a hypothetical causal mechanism that marks the progression from an exposure, through mediators, to an outcome. Conventional methods for assessing mediation effects are not applicable to a massively parallel analysis and lose statistical power when the signals are sparse. To address this, Huang (2019) proposed an adjustment procedure, which unfortunately is limited to single-mediator analyses. Our contribution is to extend the method to the setting of multiple mediators that is more realistic and commonly encountered in genomic studies. We propose a series of approaches that enables large scale multi-mediator analyses integrating Huang’s method and various multivariate non-mediation tests such as Hotelling’s t-test, variance component test, minimum p-value method, Berk-Jones test, and higher-criticism test. Our methods can be categorized into global approaches that favor dense and diverse mediation effects, and local approaches that favor sparse and consistent effects. We also provide a method-selecting guideline supported by comprehensive simulation studies. Our analysis suite has been implemented as an R package MACtest. The utility is demonstrated by an application study of The Cancer Genome Atlas Lung Adenocarcinoma data set to investigate genes and networks whose expression may be regulated by smoking-induced DNA methylation aberration.

線上參與請點選連結

附件下載