Inferring multi-locus selection in admixed populations

Admixture, the exchange of genetic information between distinct source populations, is thought to be a major source of adaptive genetic variation. Unlike mutation events, which periodically generate single alleles, admixture can introduce many selected alleles simultaneously. As such, the effects of linkage between selected alleles may be especially pronounced in admixed populations. However, existing tools for identifying selected mutations within admixed populations only account for selection at a single site, overlooking phenomena such as linkage among proximal selected alleles. Here, we develop and extensively validate a method for identifying and quantifying the individual effects of multiple linked selected sites on a chromosome in admixed populations. Our approach numerically calculates the expected local ancestry landscape in an admixed population for a given multi-locus selection model, and then maximizes the likelihood of the model. After applying this method to admixed populations of Drosophila melanogaster and Passer italiae, we found that the impacts between linked sites may be an important contributor to natural selection in admixed populations. Furthermore, for the situations we considered, the selection coefficients and number of selected sites are overestimated in analyses that do not consider the effects of linkage among selected sites. Our results imply that linkage among selected sites may be an important evolutionary force in admixed populations. This tool provides a powerful generalized method to investigate these crucial phenomena in diverse populations.Author summary: In this text we introduce a method for modeling the transition rates between local ancestry states at sites along a genome which may be in linkage to multiple sites under selection. We implement our method as a tool called AHMM_MLS which we validate with simulations and apply to admixed systems of Drosophila melanogaster and Passer italiae. In our validations we found that our tool can accurately detect the number of sites and predict their selection coefficients even when they are in linkage. We also found that the inferred selection coefficients of selected sites in linkage may be overestimated if the effects of linkage are ignored. In applying our tool to two admixed systems, we found that our lab’s previous tool AHMM-S may overestimate the number and selection coefficients of selected positions along chromosomes in these populations. Our multi-locus selection model for these admixed chromosomes produces an expected local ancestry that closely matches the local ancestry inferred from samples. We expect that our tool will allow greater insights into admixed chromosomes which show signs of a few strong selected positions that are in linkage.