Fig. 3.

The power of F_{ST_MaxSNP} is particularly sensitive to sample size. Here, the power of each statistic (y-axis) is plotted as a function of sample size (x-axis; number of chromosomes per population). We found that depending on sample size, F_{ST_MaxSNP} outperforms F_{ST_FullWin}and χ_MD for a simple isolation model, for: (A) a high N_e population with initial beneficial allele frequency of 0.005 and final frequency of 0.70, and (B) a low N_e population with initial frequency of 0.05 and final frequency of 0.80. Similar results were observed for a complete soft sweep with a population bottleneck of 0.05, except that the loss of power for F_{ST_MaxSNP} was more immediate at lower sample sizes, for: (C) a high N_e population with initial frequency of 0.05, (D) a low N_e population with initial frequency of 0.01. For partial hard sweep scenarios where F_{ST_FullWin} and χ_MD outperform F_{ST_MaxSNP}, all three statistics show more gradual sample size effects, specifically for new mutations and: (E) a final frequency of 0.40 in a high N_e population, and (F) a final frequency of 0.50 in a low N_e population.