B0420
Title: Mediation analysis with high dimensional exposures or confounders
Authors: Zhikai Yang - University of Nebraska Lincoln (United States)
Qi Zhang - University of New Hampshire (United States) [presenting]
Jinliang Yang - University of Nebraska Lincoln (United States)
Abstract: To leverage the advancements in GWAS and QTL mapping for traits and molecular phenotypes to gain a mechanistic understanding of genetic regulation, biological researchers often investigate the eQTLs that colocalize with QTL or GWAS peaks. Research is inspired by two such studies. One is in maize which aims to identify the causal SNPs that are responsible for the phenotypic variation and whose effects can be explained by their effects at the transcriptomic level. The other study in maize focuses on uncovering the cis-driver genes that lead to phenotypic changes through regulating trans-regulated genes. Both studies can be formulated as mediation problems with potentially high dimensional exposures and confounders that seek to estimate the overall indirect effect of each exposure. MedDiC, a novel procedure is proposed to estimate the overall indirect effect based difference-in-coecients approach. Simulation studies show that MedDiC offers valid inference for the indirectness with higher power than the competing methods for both low-dimensional and high-dimensional exposures. MedDiC is applied to the two aforementioned motivating datasets, and the MedDiC yields reproducible outputs across the analysis of closely related traits, and the results are supported by external biological evidence.
