Application of stochastic search variable selection to genetic association studies in common complex diseases

Abstract

We investigated the effectiveness of stochastic search variable selection (SSVS) based on Bayesian interpretation for a genetic association study in Alzheimer's disease (AD). We applied SSVS and conventional logistic regression simultaneously to two different datasets. Dataset 1 was composed of the genotypes of apolipoprotein E and choline acetyltransferase from 185 AD patients and 562 normal controls, whereas dataset 2 was composed of genotypes of apolipoprotein E and alpha-1-antichymotrypsin from 86 AD patients and 172 normal controls. Using the SSVS approach, we analyzed best final models obtained under four different prior distributions. It was found that SSVS is as powerful or more powerful than conventional logistic regression for detecting associations between the candidate genetic markers in both datasets and the presence of AD. In particular, in dataset 2, which did not contain an individual with the candidate genetic markers of interest in normal controls, SSVS did a better job than conventional logistic regression. We conclude that SSVS is a versatile tool for the analysis of multiple candidate genes with low allele frequencies simultaneously with respect to some common complex trait such as AD. © The Genetics Society of Korea.