Using step-wise linear regression to detect "functional" sequence variants: application to simulated data

Step-wise linear regression was used to detect the "functional" sequence variant in gene 6 responsible for phenotypic variation in traits Q1 and Q2. Prior to analysis, single-nucleotide polymorphisms (SNPs) that were in complete or near complete linkage disequilibrium were binned. In total, we identified 11 separate alleles (or allelic bins). Analyses were performed on all 50 replicates. The "functional" allele variant in gene 6 (at position 5782) accounted for 24% of the variation in Q1 and 11% of the variation in Q2. We detected a significant association between this SNP and Q1 in 90% of the replicates (i.e., in 45 of 50 replicates) and between this SNP and Q2 in 78% of the replicates. Although significant associations were also observed with some nonfunctional SNPs, our results nevertheless suggest that simple step-wise regression may play a useful role in analyzing sequence data. Some additional extensions to this approach are suggested.