Coalescent-based method for learning parameters of admixture events from large-scale genetic variation data

Abstract

Detecting and quantifying the timing and the genetic contributions of parental populations to a hybrid population is an important but challenging problem in reconstructing evolutionary histories from genetic variation data. With the advent of high throughput genotyping technologies, new methods suitable for large-scale data are especially needed. Furthermore, existing methods typically assume the assignment of individuals into subpopulations is known, when that itself is a difficult problem often unresolved for real data. Here we propose a novel method that combines prior work for inferring non-reticulate population structures with an MCMC scheme for sampling over admixture scenarios to both identify population assignments and learn divergence times and admixture proportions for those populations using genome-scale admixed genetic variation data. We validated our method using coalescent simulations and a collection of real bovine and human variation data. On simulated sequences, our methods show better accuracy and faster runtime than leading competitive methods in estimating admixture fractions and divergence times. Analysis on the real data further shows our methods to be effective at matching our best current knowledge about the relevant populations.