We further investigated to what extent the different platforms detect CNVs >50 kb. Results of each platform were compared at the sample level, one platform at a time, to all variants >50 kb that were identified by the other platforms. We also performed the same comparison to variants detected by at least two other platforms (Supplementary Table 6). The results of the latter analysis show that the fraction of calls missed by each platform (of the regions detected by at least two other arrays), ranges from 15% for Agilent 2×244K to ~73–77% for Illumina SNP arrays. These differences between platforms may to some extent be explained by overlap with SegDups. The Agilent 2×244K data set has a larger fraction of calls >50 kb as well as a larger fraction of calls overlapping SegDups, compared to results from the Illumina SNP arrays. Indeed, we find that 80–85% of such missing Illumina calls overlap with SegDups (Supplementary Methods). We also find that many of the calls missed by SNP arrays but detected by CGH arrays are duplications. We ascribe this effect to
