The platforms with higher resolution, as well as those specifically designed for detection of previously annotated CNVs, identified substantially higher numbers of variants compared to lower resolution arrays. The total set of CNV calls for all platforms and algorithms is given in Supplementary Table 3. To minimize the effects of outlier data on global statistics, we created a set of high-confidence CNV calls for each data set, defined as regions with at least 80% reciprocal overlap identified in at least two of the three replicate experiments for each sample (Supplementary Table 3). The size distribution of high confidence CNVs for each array and algorithm combination is shown in Figure 1 and Supplementary Figure 4a,b. Although the number of variants >50 kb is relatively even across platforms, the fraction of variants 1–50 kb in size for each platform range from >5% for Affymetrix 250K to >95% for Illumina 660W. As the arrays differ substantially in resolution and in distribution of probes, it is also relevant to investigate the distribution of probes within the CNV call made for each platform (Supplementary Fig.
