To perform the LAD we adopted ELAI24, a two-layer HMM software designed to learn the structure of local and long distance ancestral haplotypes. We adopted two combinations of 72 samples equally distributed among CEU, TSI, IBS and GUM, LWK, YRI, respectively, as sources for West-Eurasian and African segments for the Egyptian, Ethiopian and ASW samples. Concerning UKBB samples, to minimize batch effects we resorted to internally available source samples. To this extent, 100 samples were extracted from the “UK EUR” set to act as European sources, while as African and East Asian sources we extracted the 100 with the highest appropriate ancestry fraction according to ADMIXTURE. The admixture generations parameter was set as 100 for Ethiopians, 30 for Egyptians, 6 for ASW and 10 for UKBB samples. Each time we used 20 EM steps and performed 10 runs, of which we took the average. We assigned each SNP to the ancestry with the highest estimated allelic dosage, provided it reached a threshold of 0.9, or labeled it as unknown otherwise. In all cases we used phased data, treating haplotypes as independent.
