Selecting SNPs in two-stage analysis of disease association data: a model-free approach

For large numbers of marker loci in a genomic scan for disease loci, we pro pose a novel 2-stage approach for linkage or association analysis. The two stages are (1) selection of a subset of markers that are 'important' for th e trait studied, and (2) modelling interactions among markers and between m arkers and trait. Here we focus on stage 1 and develop a select ion method based on a 2-level nested bootstrap procedure. The met hod is applied to si ngle nucleotide polymorphisms (SNPs) data in a cohort study of heart diseas e patients. Out of the 89 original SNPs the method selects 11 markers as be ing 'important'. Conventional backward stepwise logistic regression on the 89 SNPs selects 7 markers. which are a subset of the 11 markers chosen br o ur method.