A Bayesian Regression Model with Variable Selection for Genome-Wide Association Studies
In the previous chapter, we have seen the use of stochastic searching for variable selection in a simple linear model setting. In this chapter, we extend the concept for a specific application, namely genome-wide association studies (GWAs). GWAs aim to identify, from among a large number of marker loci drawn from across the genome, those markers that are in linkage disequilibrium with a locus associated with some disease or phenotypic trait. Due to increasing knowledge of common variations in the human genome, advancements in genotyping technologies and in particular reduction in the cost of gene chips, GWAs have become more prevalent. One of the current challenges faced in GWAs is to find an adequate and efficient statistical method for analysing large single nucleotide polymorphism (SNP) data sets.
In this chapter, we propose to regress the trait or disease status of interest against all SNPs simultaneously with the stochastic variable selection algorithm. Excellent methods for the variable selection problem have been developed within a Bayesian context. The issue of multiple comparisons is also handled simply and effectively in a Bayesian context (Berry and Hochberg ...