We started by assessing the accuracy of ELAI24, a local ancestry inference (LAI) or deconvolution (LAD) software, in simulated scenarios resembling our studied populations. We tested several combinations of admixtures between deeply divergent populations (i.e that trace their ancestry to East Asia, Africa and Europe) and number of generations elapsed since the admixture (Supplementary Fig. 2), reporting in all cases a classified accuracy equal or higher than 0.98 when using 0.9 as the minimum inferred ancestry dosage. We then subdivided the real admixed whole genome sequences of individuals with both African and West-Eurasian background into their ancestral components through this process. This gave us two genomic portions, one African- and one European-related, together with an unassigned fraction for each genome (Supplementary Fig. 3).
