Although detection of variants in the 1–50 kb range is important for research and discovery projects, clinical laboratories generally remove or disregard smaller variants. To address the question of reproducibility across different size ranges, all CNVs were divided into size bins, and the replicate reproducibility was analyzed in each bin. We performed this analysis separately for the different algorithms, platforms and sites (Supplementary Table 4a–c, respectively). Contrary to our expectations, we found that reproducibility is generally similar for large and small CNVs. We note that the reproducibility of large CNV calls is disproportionally affected by genome complexity as they tend to overlap SegDups to a larger extent than small CNVs: 55% of nonreplicated large calls (>200 kb) have at least 50% overlap with SegDups, compared to 4% for small calls (1–10 kb) (Supplementary Table 5). SegDups tend to complicate probe design, suffer from low probe coverage and cross-hybridization, and they are therefore often refractory to CNV detection using array-based techniques. Indeed, CNVs overlapping SegDups generally have fewer probes supporting them (Supplementary Fig. 7) and their reproducibility is lower compared to
