Multiple-Ancestor Localization for Recently Admixed Individuals

Inference of ancestry from genetic data is a fundamental problem in computational genetics, with wide applications in human genetics and population genetics. The treatment of ancestry as a continuum instead of a categorical trait has been recently advocated in the literature. Particularly, it was shown that a European individual’s geographic coordinates of origin can be determined up to a few hundred kilometers of error using spatial ancestry inference methods. Current methods for the inference of spatial ancestry focus on individuals for whom all ancestors originated from the same geographic location. In this work we develop a spatial ancestry inference method that aims at inferring the geographic coordinates of ancestral origins of recently admixed individuals, i.e. individuals whose recent ancestors originated from multiple locations. Our model is based on multivariate normal distributions integrated into a two-layered Hidden Markov Model, designed to capture both long-range correlations between SNPs due to the recent mixing and short-range correlations due to linkage disequilibrium. We evaluate the method on both simulated and real European data, and demonstrate that it achieves accurate results for up to three generations of admixture. Finally, we discuss the challenges of spatial inference for older admixtures and suggest directions for future work.