Integrating common and rare genetic variation in diverse human populations.

Size and frequency spectra of common and rare CNPsa, Estimated size distribution of common CNPs calculated from the physical span of the genomic probes supporting each CNP event. b, Allele frequency spectrum for biallelic CNPs calculated from integer CNP genotypes for the samples analysed in this work.

SNP discovery informativeness across populationsa, b, For each of 7 populations for which at least 60 individuals were resequenced, we considered a sample of 30 individuals, another non-overlapping sample of 30 individuals from the same population, and a sample of 30 individuals from each of the 6 other populations (results are averaged over 1,000 random samplings). Out of all SNPs that are either polymorphic (a) or polymorphic with a minor allele with at most two copies in the sample of 30 individuals (b), here we present the fraction that are also polymorphic in a different sample, starting with the other sample from the same population (black bars). The black bars serve as a baseline that accounts for the effect of sampling stochasticity and sequencing errors on SNP discovery. The different y-axis scales used reflect the lower likelihood of a low-frequency variant being seen in a different sample.

Effect of sample size on SNP ascertainmentThe number of SNPs discovered as a function of sample size by averaging over 1,000 random samplings. For each population, we randomly sampled without replacement a subset of the individuals of any possible size and considered which SNPs were polymorphic in the resequencing data for that sample. For any given sample size, many more variants are discovered in populations with genetic proximity to Africa (LWK, ASW and YRI), compared to populations of non-African ancestry.

Haplotype sharing around SNPs and CNPsa, b, Extent of haplotype homozygosity around variant alleles of various frequencies. Shown are SNPs from the ENCODE sequence, CNPs of comparable frequency, SNPs from the arrays and on randomly grouped chromosomes, and (for YRI) the maximum possible sharing for a genotyping error rate of 0.2%. a, CEU. b, YRI.

Imputation accuracy and reference panel sizea, b, Mean r2 between true and imputed genotype dosage for SNPs imputed from a HapMap-II-sized panel of 120 CEU chromosomes (HMII-CEU) or a HapMap 3 panel of 410 European-ancestry chromosomes (CEU+TSI). Scatter plots show Affymetrix 500K SNPs on chromosome 20 imputed for 1,393 subjects of the 1958 British birth cohort. a, Rare SNPs (MAF <0.5%). b, Low-frequency SNPs (MAF = 0.5–5%).

Imputation: new populations, new variantsa, b, Mean r2 between true and imputed genotype dosage as a function of copies of minor allele in the reference panel. a, The loss in imputation accuracy when the reference population differs slightly from the target population (CEU imputed into CEU compared to CEU into TSI; and YRI into YRI compared to YRI into LWK). b, Imputation accuracy for newly discovered variants (CNPs and ENCODE SNPs).