High-Dimensional Mediation Analysis with Applications to Causal Gene Identification

Abstract

Mediation analysis has been a popular framework for elucidating the mediating mechanism of the exposure effect on the outcome. Previous literature in causal mediation primarily focused on the classical settings with univariate exposure and univariate mediator, with recent growing interests in high dimensional mediator. In this paper, we study the mediation model with high dimensional exposure and high dimensional mediator, and introduce two procedures for mediator selection, MedFix and MedMix. MedFix is our new application of adaptive lasso with one additional tuning parameter. MedMix is a novel mediation model based on high dimensional linear mixed model, for which we also develop a new variable selection algorithm. Our study is motivated by the causal gene identification problem, where causal genes are defined as the genes that mediate the genetic effect. For this problem, the genetic variants are the high dimensional exposure, the gene expressions the high dimensional mediator, and the phenotype of interest the outcome. We evaluate the proposed methods using a mouse f2 dataset for diabetes study, and extensive real data driven simulations. We show that the mixed model based approach leads to higher accuracy in mediator selection and mediation effect size estimation, and is more reproducible across independent measurements of the response and more robust against model misspecification. The source R code will be made available on Github https://github.com/QiZhangStat/highMed upon the publication of this paper..

Multimodal data integration via mediation analysis with high‐dimensional exposures and mediators

“…Recently, Aung et al (2020), Long, Irajizad, Doecke, Do, and Ha (2020), and Zhang (2021) proposed new approaches for mediation analysis of multivariate exposures and mediators. In particular, Zhang (2021) developed two regularization procedures and applied them to a mouse f2 dataset for diabetes, taking SNP genotypes as the exposures, islet gene expressions as the mediators, and insulin level as the outcome. However, they required the mediators to be independent, which hardly holds in our setting, as different brain regions are generally believed to influence each other.…”

“…While there have been numerous extensions of mediation analysis to account for multiple mediators (see, e.g., Chén et al, 2017; Song et al, 2018; Zhao & Luo, 2022, among many others), there have been very few works studying multivariate exposures, or both multivariate exposures and mediators. Recently, Aung et al (2020), Long, Irajizad, Doecke, Do, and Ha (2020), and Zhang (2021) proposed new approaches for mediation analysis of multivariate exposures and mediators. In particular, Zhang (2021) developed two regularization procedures and applied them to a mouse f2 dataset for diabetes, taking SNP genotypes as the exposures, islet gene expressions as the mediators, and insulin level as the outcome.…”

Multimodal Data Integration via Mediation Analysis with High-Dimensional Exposures and Mediators

“…While there have been numerous extensions of mediation analysis to account for multiple mediators (see, e.g., Zhao and Luo, 2016;Chén et al, 2017;Song et al, 2018, among many others), there have been very few works studying multivariate exposures, or both multivariate exposures and mediators. Recently, Zhang (2019); Aung et al (2020); Long et al (2020) proposed new approaches for mediation analysis of multivariate exposures and mediators. In particular, Zhang (2019) developed two regularization procedures and applied them to a mouse f2 dataset for diabetes, taking SNP genotypes as the exposures, islet gene expressions as the mediators, and insulin level as the outcome.…”

“…Recently, Zhang (2019); Aung et al (2020); Long et al (2020) proposed new approaches for mediation analysis of multivariate exposures and mediators. In particular, Zhang (2019) developed two regularization procedures and applied them to a mouse f2 dataset for diabetes, taking SNP genotypes as the exposures, islet gene expressions as the mediators, and insulin level as the outcome. However, they required the mediators to be independent, which hardly holds in our setting, as different brain regions are generally believed to influence each other.…”

“…Additional simulations further demonstrate the efficacy of the method. Similar to Zhang (2019); Aung et al (2020); Long et al (2020), our approach is among the first attempts to conduct mediation analysis where both the exposures and mediators are high-dimensional. But unlike the existing solutions, we do not restrict the dimensionality or the correlation structures and do not require additional domain knowledge of the exposures or mediators.…”