Random sets of genes are also employed to estimate the study wide p-value by applying a bootstrap method: one of the 50,000 sets of genes is selected as a reference and compared to a subsample of 5,000 random lists (selection with replacement). This is repeated 1,000 times, comparing the most significant p-value in each iteration to the p-value of each category of genes, estimating how often the significance level is obtained by random chance. In addition, the number of sets of genes that achieved different threshold levels (i.e., <0.005; 0.01 and 0.05) is compared to the values obtained by random reference studies to calculate the excess of significantly overrepresented sets of genes [15]. The method can correct for any bias due to genes that are physically located within the same region of the chromosome (genes which are less than 1 Mb apart) associated with the same signal and assigned to the same category of genes. This is done by grouping these genes as a single entity, which is tagged by a set of SNPs mapping within the genes.
