Genome-Wide eQTL Mapping by Iterative Multivariate Adaptive LASSO (LMAL)

• Activity Number: 133
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 8:30 AM to 10:20 AM
• Sponsor: ENAR
• Author(s): Ting-Huei Chen*+ and Wei Sun and Fred Andrew Wright
• Companies: The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill and The University of North Carolina
• Keywords: Multivariate penalized regression ; iterative adaptive Lasso ; multiple loci mapping ; gene expression QTL ; variable selection

Abstract

Genome-wide eQTL (gene expression quantitative trait loci) mapping aims to select important genetic markers that explain the variances of the gene expressions. Therefore, it can be treated as a variable selection problem with multiple responses and covariates. Many methods have been proposed to analyze each gene expression trait separately. However, it is well known that several genes could be co-regulated and linked to some common genetic markers. The trait-by-trait mapping strategy fails to take into account the possible correlation between traits, which may lead to a loss of power. We propose IMAL, a multivariate penalized regression method, for eQTL mapping. IMAL employs two penalties to fulfill the parsimonious model assumption and accommodate possible genetic hotspots. Moreover, IMAL incorporates the correlation among traits and is applicable for the high dimension low sample size setting. Asymptotic studies show that the penalty function of IMAL has better capability to handle high dimensional problem than most popular penalties. Empirical results from both simulation and real data analysis confirm that IMAL has improved variable selection performance than existing methods.