Population divergence time estimation using individual lineage label switching

Divergence time estimation from multilocus genetic data has become common in population genetics and phylogenetics. We present a new Bayesian inference method that treats the divergence time as a random variable. The divergence time is calculated from an assembly of splitting events on individual lineages in a genealogy. The time for such a splitting event is drawn from a hazard function of the truncated normal distribution. This allows easy integration into the standard coalescence framework used in programs such as Migrate. We explore the accuracy of the new inference method with simulated population splittings over a wide range of divergence time values and with a reanalysis of a dataset of 5 populations consisting of 3 present-day populations (Africans, Europeans, Asian) and 2 archaic samples (Altai and Ust'Isthim). Evaluations of simple divergence models without subsequent geneflow show high accuracy, whereas the accuracy of the results of isolation with migration models depends on the magnitude of the immigration rate. High immigration rates lead to a time of the most recent common ancestor of the sample that, looking backward in time, predates the divergence time. Even with many independent loci, accurate estimation of the divergence time with high immigration rates becomes problematic. Our comparison to other software tools reveals that our lineage-switching method, implemented in Migrate, is comparable to IMa2p. The software Migrate can run large numbers of sequence loci (>1,000) on computer clusters in parallel.