Local Ancestry Inference for Complex Population Histories

Abstract

It has become apparent from ancient DNA analysis, that the history of many human populations from across the globe are often complex, involving multiple population split, admixture, migration and isolation events. Local ancestry inference (LAI) aims to identify from which ancestral population chromosomal segments in admixed individuals are inherited. However, ancestry in existing LAI tools is characterised by a discrete population identity, a definition which is limited in the context of a complex demographic history involving multiple admixture events at different times. Moreover, many LAI tools rely on a reference panel of present day genomes that act as proxies for the ancestral populations. For ancient admixture events, these proxy genomes are likely only distantly related to the true ancestral populations. Here we present a new method that leverages advances in ancient DNA sequencing and genealogical inference to address these in issues in LAI. The method applies machine learning to tree sequences inferred for ancient and present day genomes and is based on a deterministic model of population structure, within which we introduce the concept of path ancestry. We show that the method is robust to a variety of demographic scenarios, generalises over model misspecification and that it outperforms a leading local ancestry inference tool. We further describe a downstream method to estimate the time since admixture for individuals with painted chromosomes. We apply the method to a large ancient DNA dataset covering Europe and West Eurasia and show that the inferred admixture ages are a better metric than sample ages alone for understanding movements of people across Europe in the past.