We also measured breakpoint accuracy at the sample level by comparing deletion calls in this study with deletions from the 1000 Genomes Project45,46. Four of the samples used in this study were sequenced by that project, and those samples had 3,124–4,200 breakpoints annotated. The data were compared at the sample level for each combination of platform and algorithm. The results for a representative sample (NA18517) are displayed in Figure 4b (remaining samples in Supplementary Fig. 14), showing the overlap and breakpoint distance for each breakpoint from the 1000 Genomes Project. The results of these analyses are similar to those above, where all platforms show underestimation of the variable regions. Compared to other platforms the CNV-enriched SNP arrays (Illumina 660W and Omni) detect a substantially larger number of variants, which are present in the data from the 1000 Genomes Project. The Agilent 2×244k array, which also targets known CNVs, performs extremely well in relation to its probe density, especially when analyzed with the ADM-2 algorithm.
