In all cases, FST, a measure of the degree of population differentiation (Supplementary Table 6) correctly predicted the most informative population, despite the FST estimates being based on genotyping array data with SNP ascertainment biases15. However, FST was not a perfect predictor: the correlation coefficient between FST and ascertainment informativeness was highly variable, ranging across populations between −0.67 and −0.99 for all SNPs and between −0.51 and −0.97 for low-frequency SNPs. Furthermore, FST is symmetrical between a pair of populations, whereas informativeness is not. For example, the most informative population for low-frequency GIH SNPs was TSI, with informativeness being only 55% of that of an independent GIH sample (because TSI captures only one of GIH’s ancestral populations; Fig. 2b). Conversely, the informativeness of GIH on low-frequency TSI SNPs was 71% (Fig. 2b).
