Two alternative approaches for measuring the effect of ascertainment bias on admixture graph fits are illustrated using one population combination, “Denisovan, Khomani San, Mbuti, Dinka, Mursi”.

1) residual standard deviation (residual SE) of linear trends and 2) squared Pearson correlation coefficient (R2) for two admixture graph fit metrics (worst f4-statistic residuals, WR, or log-likelihood scores, LL) calculated on unascertained vs. ascertained data. Five thousand best-fitting graphs (according to LL on all sites) of 32,745 possible graphs were selected, and correlation of LL (left-hand panels) or WR (right-hand panels) was explored for graphs fitted on all sites and on ascertained datasets. Results for ascertainment on variants common in Africans (either those having no detectable West Eurasian ancestry or all Africans in the SGDP dataset) are circled in red. As a starting point for generating different ascertained datasets, we used 11,706,773 sites (with no missing data at the group level) polymorphic in a set of 48 archaic and African groups composed of 97 individuals (S1 Table). Thirty-eight site subsampling schemes were explored (see a list in the legend for Fig 2). The size of the resulting SNP panels is coded by point size, and ten broad ascertainment types are coded by color according to the legend in the upper right corner. The 97.5th (in the case of residual SE) or 2.5th (in the case of R2) LL or WR percentiles of all the thinned replicates combined, including those on all sites and AT/GC sites, are marked by brown lines. Areas of the plots where ascertainments are considered biased according to these thresholds are highlighted in red on the left-hand side of the plots. Scatterplots illustrating effects of selected ascertainment schemes (marked with numbers 1 to 7) on LL or WR are shown in the middle of the figure. Each dot on these scatterplots corresponds to a distinct admixture graph topology. (PDF)