Figure 2.

RMSD (left) and mean bias (right) for maximum-likelihood estimates of F_ST under different sequencing coverage (2×, 6×, and 20×). We compared accuracy of the new method, which does not rely on genotype calling $\left({\widehat{F}}_{\text{ST.ML}}\right)$, and the standard method applied to called genotypes. $\left({\widehat{F}}_{\text{ST.ML.GC}}\right)$ (see Materials and Methods). We simulated 20 individuals for each population and 1000 sites for each scenario.