Unlike SNPs present on microarray platforms, which are intentionally biased towards high frequency by the discovery and selection process, the SNPs discovered by sequencing provide a direct estimate of the underlying allele frequency spectrum in each population. As in previous surveys, common (MAF ≥5%) and low-frequency (MAF = 0.5–5%) variants account for the vast majority of the heterozygosity in each sample, but we also observed a large number of rare (MAF = 0.05–0.5%) and private (singletons and MAF <0.05%) variants (see Supplementary Table 2 for definitions of variant frequency classes). Each population had 42– 66% of sites with a MAF <5%, compared to 10–13% in the genotyping data; 37% of SNPs with a MAF <0.5% were observed in only one population. In total, 77% of the discovered SNPs were new (that is, not in the SNP database (dbSNP) build 129) and 99% of those had a MAF <5%.
