Fig. 4.

ROC curves for binary classification of selective sweeps, including partial sweeps, versus regions neutrally evolving or under linked selection. Given the BFS population history. Each curve is labeled with a number that is indexed in the legend. The partialS/HIC deep learning classifier outcompetes two other approaches for managing the same suite of summary statistics: Composite of Multiple Signals and PCA. Furthermore, partialS/HIC excels in performance over several subwindow derivatives of two summary statistics, SAFE and iHS, which were all included within our set of summary statistics used for training. Notably, Composite of Multiple Signals, SAFE, and iHS were all designed to uncover selective sweeps. Additionally, the SAFE score is itself a compound statistic that captures signal from several constituent statistics, and the majority of our training data is derived from either the SAFE score or one of its components.