Figure 5
Estimated error in the estimated individual admixture proportions from the simulated admixed population (HI model). We used an extended version of the backward demographic simulator described in [76] that includes recombination and different types of mating and allows for ancestry painting [14]. Over all parameters that are defined in this model [19], we varied the time of split of the ancestral populations, which ranged between 50 and 2,000 generations among simulations. Each simulation generated 75 (25 by population) full human genomes with 22 diploid chromosomes (l) with the following sizes: 13.65, 13.15, 11.20, 10.65, 10.20, 9.65, 9.35, 8.50, 8.40, 8.95, 7.95, 8.65, 6.35, 5.80, 6.30, 6.75, 6.50, 5.95, 5.40, 5.40, 3.10, and 3.65 Mb [77]. The mutation rate was set to 2.5 × 10−8 [78] and the recombination rate to 1.8 × 10−8. PLINK was applied to exclude SNPs with minor allele frequency less than 0.05 and LD (default PLINK --indep 50 5 2). The effective population sizes of the parental and hybrid populations were set to 5,000 diploid individuals; the time of admixture was ten generations ago, and each parental population equally contributed to the admixed population. By this way, we minimized the putative effect of genetic drift in the admixture proportions of the hybrid population. Furthermore, in order to include the effects of bias sample size, we repeated all the analyses with 1:1 (A) and 1:5 (B) parental population size ratios. Four different algorithms were considered: sNMF, ADMIXTURE, fastSTRUCTURE, and MDS. In the case of MDS, ancestry proportions of each individual from the admixed population were estimated as the relative position in the first dimension in relation to the mean estimated coordinate of the parental populations.