Inference of complex demographic history using composite likelihood based on whole-genome genealogies

Abstract

Accurate parametric inference on complex demographic models is a continuing challenge in population genetics. Ancestral recombination graphs (ARGs) provide richer information than simple population genetic summary statistics and can potentially improve the power and accuracy of such inference. We present mrpast, a tool for inferring complex demographic history from ARGs. mrpast uses a composite likelihood formulation based on the pairwise sample coalescence time, observable in an ARG, and the coalescence probabilities from a continuous-time Markov process. We have evaluated mrpasts accuracy on a variety of models, including stepping-stone models with asymmetric migrations, changes in effective population sizes, out-of-Africa, and American admixture. We demonstrated mrpasts accuracy using simulated ARGs and inferred ARGs, and its high versatility in jointly inferring all parameters in complex models, including time of demographic events (e.g., population split, admixture), effective population size (e.g., constant, exponential growth), and gene flow (e.g., admixture proportion, migration rate). Extending the three-population out-of-Africa model with asymmetric migrations, we observed significantly more migrations from East Asians to Europeans than from Europeans to East Asians. Notably, mrpast can reliably recover all parameters in an American admixture model, when treating non-admixed Native Americans as an unsampled ("ghost") population. Applying this model to Mexican, Puerto Rican, and Colombian populations, we found that the reconstructed histories of Native and admixed Americans align closely with both historical records and genetic evidence. Lastly, mrpast provides a comprehensive pipeline to facilitate easier, more appropriate, and robust demographic inference, in which users can easily simulate, infer, and manipulate ARGs, and illustrate and test a demographic model.