We then assayed 450 HapMap samples (180 CEU, 180 YRI, 45 JPT (individuals in Tokyo, Japan) and 45 CHB (individuals in Beijing, China)) across our CNV-typing array. We used a Bayesian algorithm to genotype CNVs (more precisely: to assign individuals to diploid copy number classes), and then manually curated the selection of the optimal normalization and cluster locations for every locus (Supplementary Methods). We applied quality-control filters to identify 5,238 non-redundant CNVs (4,978 from the CNVs discovered here) that could be genotyped with high confidence in at least one HapMap population (3,320 were polymorphic in CEU, 3,985 in YRI and 1,957 in JPT+CHB), and these genotypes exhibited high concordance across replicate experiments (Supplementary Table 2 and Supplementary Methods).
