We obtained CNV calls from the raw hybridization intensities using PennCNV [19]. We excluded from this analysis samples that were outliers based on either the variability of the raw intensity data (using the standard deviation of the logR ratio), or on the total number of CNVs called (see Methods S1 and Figure S2). This resulted in 413 samples being retained for further analysis (179 cases and 234 controls). To make the final CNV calls, we used the following criteria: (i) we merged neighbouring CNVs when the distance separating them was less than half of the total distance from the start of the first CNV to the end of the second CNV, (ii) we only called CNVs containing at least 10 SNPs, and (iii) we ignored CNVs located in centromeric and telomeric regions.
