Comparison of estimates of admixture proportion.

The proportion of admixture for each of the admixed populations pooled in this study was estimated using a regression-based method (see Methods). Wherever possible, we also estimated the proportion of admixture using genotypes at AIMs known to distinguish the HapMap populations. βYRI, βCEU, and βCHB/JPT are the regression coefficients, which are taken as the proportion of ancestry contributed by each of the YRI, CEU, and CHB/JPT populations. The standard error (s.e.) of the regression coefficient is also listed when available. Note that the s.e. may be biased downward, due to LD between SNPs. However, the s.e. estimates based on the LD-pruned set of SNPs are on the order of 10−3 (data not shown). Intercept is the regression intercept, which in this case is half of the unexplained ancestry in the model, as the average allele frequency for the population of interest and each of HapMap populations is ∼0.5 (Text S2). NSNP is number of SNPs used to generate the ancestry estimates (see Methods). The “method” column indicates the method used to generate the admixture estimates: “pooling” indicates that estimates are based on regression from pooled allele frequencies, “genotype” indicates that estimates are based on regression from individual genotype data, and “AIMs” indicates that estimates are were generated using STRUCTURE and individual genotype data from a small number of AIMs (see Text S1). n.d. denotes not determined; n.a. denotes not available. Estimates based on the regression approach do not appear to be confounded by issues due to collinearity (data not shown). For MAY, GXE, and SPT, when SNP allele frequencies from CHB/JPT were included in the model, βYRI was be largely unchanged, but a small contribution (<0.025) from CHB/JPT was estimated. This small admixture contribution from CHB/JPT appears to be largely an artifact due to sampling variation of the CEU and CHB/JPT reference populations (C.W.K.C., unpublished).