Mediation Analysis with Missing Data for Genomics

• Activity Number: 181 - Statistical Methods in Gene Expression Data Analysis II
• Type: Contributed
• Date/Time: Tuesday, August 4, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Section on Statistics in Genomics and Genetics
• Author(s): Won Gyo Suh* and Fred A. Wright and Yi-hui Zhou
• Companies: Statistics , NCSU and Statistics, Biological Sciences, and Bioinformatics Research Center. NCSU and Biological Sciences, Statistics, and Bioinformatics Research Center. NCSU
• Keywords: statistical genetics; genomics; mediation analysis; missing data

Abstract

Mediation analysis has become a standard approach to elucidate the joint effects of genotype and other ‘omics platforms on phenotypes, with recent efforts focused on effectively combining sources of high-dimensional data. However, a large proportion of missing data are common in such studies and providing efficient estimators can be challenging. General Monte-Carlo simulation approaches to handle missing data patterns are computationally challenging, even when data are missing completely at random. Also, multiple imputation can be dependent on the choice of the imputation model and may be less statistically efficient than maximum likelihood. In genomics settings, only potential mediators such as gene expression or other ‘omics data have a high proportion of missing data. We demonstrate that such studies offer the potential to explicitly maximize the likelihood in a computationally feasible manner. We provide indirect mediator effect tests using estimators from the model. We illustrate using simulations and using data from genome-wide association studies for combinations of data from single-nucleotide polymorphisms and expression data from multiple tissue sources.