We performed the same analysis for CNPs, studying variants in the same frequency range (two to six copies) and in the same two populations. To reduce ambiguity, we restricted ourselves to CNPs that had exactly two genotypic states and treated these as biallelic variant sites. We masked out any SNPs within the boundaries of the CNP, and thereafter analysed them in the same way as the SNPs. We found that CNPs and SNPs in the same samples had a similar extent of haplotype sharing (Fig. 4); in CEU, sharing does drop off faster for CNPs, but the difference was not statistically significant with our sample size. This observation, which is consistent with a previous observation that low-frequency CNPs segregate on long shared haplotypes7, suggests that imputation methods should have comparable effectiveness for CNPs as for SNPs, at least for biallelic CNPs that are measured well by our array-based approach.
