2 | MATERIALS AND METHODS
We evaluated how Approximate Bayesian Computation model-choice and posterior parameter estimation performed for reconstructing highly complex historical admixture processes from genetic data. To do so, we chose to work under the two source-populations version of the general mechanistic model of Verdu and Rosenberg (2011) briefly presented inSupplementary Figure S1 . We introduce a novel software,MetHis , for genetic data simulation and summary-statistics calculation for machine-learning ABC inferences under this general model (Supplementary Note S1 ).
We conduct our proof of concept considering nine competing scenarios of complex admixture histories involving multiple admixture pulses, recurring decreasing or increasing admixture, and combinations of these processes (Figure 1 , Table 1 ). We explore the recent admixture history of two enslaved-African descending populations in the Americas with genome-wide independent SNPs. Beyond this work, theMetHis -ABC framework can readily be used to study numerous histories of complex admixture using independent SNP or microsatellite markers (Supplementary Note S1 ).