Although the selection algorithm used the P-value data for 550 K SNPs from the CGEMS GWAS, we did not, in this example, use information from other GWAS data sets or from the validation portion of the CGEMS initial study (4). The CGEMS follow up study was particularly robust because it genotyped 26 958 SNPs, including all SNPs with P-value <0.068 from the initial CGEMS GWAS, in 3941 cases and 3964 controls (5). This provides an unbiased opportunity to evaluate whether the very small set of SNPs selected by our algorithm include the SNPs validated in a genotyping panel that was many times larger. The CGEMS validation study identified seven prostate cancer related SNPs which had P-value ranks in the initial GWAS ranging from 116 (P = 0.0004) to 24407 (P = 0.042). Our algorithms selected five (71%) of those seven SNPs. Three of the five SNPs were selected by GenePipe, one was selected by GenomePipe and three were selected by LinkagePipe. Of the two SNPs that were missed, rs10486567 in JAZF1 was not in our candidate gene list because at
