values of non-admixed European and African populations. This cannot be attributed to a casual fragmentation of the genome: shifts towards or away from 0 are found significant in most cases when comparing with simulated aspPSs distributions obtained assigning a random local ancestry pattern (one-sided Wilcoxon signed-rank test, see Fig. 2, Supplementary Data 1). However, decreasing the size of the genomic portion p, used in computing aspPS, also decreases the directional bias following a squared root relation with the forenamed genomic fraction p. We can eliminate this effect by correcting the aspPS for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1/\sqrt{p}$$\end{document}1/p, at the expense of an increased standard deviation for small genomic fractions. Nevertheless, including these “corrected” aspPS in parameter fitting proved to disrupt trait predictability, thus suggesting to bypass this correction. See Supplementary Note 1 for a more detailed description of this correction and related data.
