Sparse Mediation Analysis Using Mixture Models

• Activity Number: 237 - SPEED:Statistical Methods for GWAs, Genetics, Genomics, and Other Omics Studies, Part 1
• Type: Contributed
• Date/Time: Monday, July 29, 2019 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Yanyi Song* and Xiang Zhou and Min Zhang and Wei Zhao and Yongmei Liu and Sharon Kardia and Ana Diez Roux and Belinda Needham and Jennifer Smith and Bhramar Mukherjee
• Companies: University of Michigan and University of Michigan and University of Michigan and University of Michigan and Wake Forest School of Medicine and University of Michigan and Drexel University and University of Michigan and University of Michigan and University of Michigan
• Keywords: Bayesian sparse models; Composite null; Epigenetics; High-dimensional mediators

Abstract

Causal mediation analysis aims to examine the role of a mediator or a group of mediators in the pathway between an exposure and an outcome. Recent biomedical studies often involve a large number of potential mediators based on high-throughput technologies. With the rapid growth of multi-platform omics data, it is desirable to perform joint analysis of molecular-level genomics data with epidemiological data through mediation analysis. However, such joint analysis requires statistical methods that can simultaneously accommodate high-dimensional and potentially correlated mediators and that are currently lacking. To bridge this gap, we develop a four-component Gaussian mixture model in a high-dimensional setting for imposing a sparsity inducing joint prior on the mediator-exposure and outcome-mediator coefficients. We compared our method to two existing approaches: two-stage Lasso, which penalizes the mediation regression models separately, and Pathway Lasso, which directly penalizes the indirect effect. We show that our method is more powerful in identifying true non-null mediators and provides accurate estimates of the proportion of mediators falling in each of the four components.