To estimate the accuracy of CNV calls, we compared the variants from each array and algorithm to existing CNV maps (Fig. 2c). We used four different `gold standard' data sets for comparison to minimize potential biases (Supplementary Fig. 10). The first data set is based on the Database of Genomic Variants (DGV v.9). We downloaded all variants in DGV and filtered the data to yield a high-quality gold standard data set (Supplementary Methods). The second data set we used was 8,599 validated CNV calls from a tiling-oligo CGH array from the Genome Structural Variation consortium11. The same study also produced CNV genotype data for 4,978 variants in 450 HapMap samples, including five of the six samples used in the present study (for sample-level comparisons, see Supplementary Fig. 11). Finally, we also used data from paired-end mapping based on fosmid end sequencing41.
