We next characterized the extent to which alleles share haplotype backgrounds as a function of frequency, a question related to the imputation of variants not directly observed in each clinical sample. Population genetic models predict that lower-frequency variants should on average be younger than more common variants, and thus have a longer physical extent of haplotype sharing. We selected from the ENCODE data a set of SNPs observed two to six times in YRI or in CEU; we estimated haplotype phase with high confidence using parent–offspring trio data. After validation using Sequenom genotyping (Supplementary Methods) to ensure highly accurate genotypes, 272 SNPs were examined in YRI and 106 in CEU. For comparison, a set of SNPs from the genotyping arrays with the same frequencies were analysed. Haplotype sharing was measured by calculating the haplotype homozygosity (that is, perfect concordance between haplotypes) using the consensus genotype data around each low-frequency SNP.
