We classify only 7 of neutrally evolving regions as sweeps. Thus, while these 3 solutions all have comparable sensitivity to sweeps in this situation, S/HIC has superior specificity: SFselect+ and evolBoosting+ classify a large fraction of unselected regions (like each neutral and sweep-linked regions) as sweeps, whereas S/ HIC features a low false optimistic price. For the curious reader, we present S/HIC’s function rankings for classifiers educated on both human demographic histories in S2 Table. Next, we examined the impact of demographic misspecification on energy to detect selection occurring under Tennessen et al.’s model in the population size history of Europeans following their migration out of Africa [44] but possessing educated S/HIC below the typical neutral model. This demographic history presents an even greater challenge for identifying positive choice than the African model, because it is characterized by two population contractions followed by exponential growth, then a a lot more recent phase of more quickly population development (Techniques). For this situation, a single selection of selection coefficients was utilised: U(5.003, 5.005). Right here, we find that, possibly unsurprisingly, the overall performance of most strategies is decrease than within the African situation. Nonetheless, S/HIC after once again appears substantially far more robust to misspecification of the demographic model than other solutions (AUC = 0.8127 versus 0.7250 for evolBoosting +, and 0.six or less for all other strategies; Fig six). Next, we examined the proportion of windows at many distances from sweeps that are assigned to every class below this situation of demographic misspecification. We discover that even though S/HIC classifies hard sweeps with lower sensitivity than under constant population size situation (56.0 and 19.1