Practical issues for screening and variable selection method in a Genome-Wide Association Analysis

Abstract

Variable selection plays an important role in high dimensional statistical modeling analysis. Computational cost and estimation accuracy are two main concerns for statistical inference of high dimensional data. Recently, many high dimensional data have been generated in biomedical science such as microarray data and single nucleotide polymorphism (SNP) data. Especially, the genome-wide association studies (GWAS) which focus on identifying SNPs associated with a disease of interest, have produced ultra-high dimensional data. Numerous methods have been proposed to handle GWAS data. Most statistical methods have adopted a two-stage approach: (1) pre-screening for dimensional reduction, (2) variable selection for identification of causal SNPs. The pre-screening step selects SNPs in terms of their p-values or absolute value of regression coefficients in single SNP analysis. Penalized regression such as Ridge, Lasso, adaptive Lasso and Elastic-net are commonly used for the variable selection step. In this paper, we investigate which combination of prescreening method and penalized regression performs best on continuous type response variable via real GWA data containing 327,872 SNPs from 8842 individuals.