In both populations, ENCODE variant alleles had longer shared haplotypes than array-based SNPs of the same frequency, and all low-frequency alleles (whether or not discovered by sequencing) had longer haplotypes than did higher-frequency SNPs (Fig. 4). Shorter haplotypes for array SNPs are expected because of SNP ascertainment, which was biased towards SNPs shared across populations and therefore towards older SNPs with shorter-range linkage disequilibrium. Among the ENCODE SNPs, there was little difference in haplotype sharing between alleles seen twice and those seen four to six times in the sample, indicating that these minor differences in frequency are not good predictors of the age and haplotype sharing of alleles (presumably due to drift and sampling error in the frequency estimate). Haplotype sharing was also greater for derived than for ancestral alleles, although the effect was modest (Supplementary Fig. 6).
