In order to allow for the effect of differing sample size we used a simulation procedure to generate the distribution of the number of hits for subsets of the sample with bipolar disorder v. controls under the null hypothesis. We randomly selected samples of individuals with bipolar disorder, where the number of individuals is the same size as the diagnostic set of interest, and used each in a case–control genome-wide analysis against the total set of controls. For each of the subsets we undertook a genome-wide analysis as described above (including both genomic control adjustment and filtering for independent signals) and counted the number of SNPs that exceeded the threshold of significance (P<10–5). This whole procedure was repeated 1000 times to produce a distribution of the number of SNPs expected by chance when testing individuals with bipolar disorder v. controls.
