Inferring demographic history from genomic data

In the last two decades, several methods to infer the demographic history of a population from whole genome sequence data have been developed. Despite these efforts, current methods remain computationally demanding, limiting researchers’ ability to efficiently explore parameter space and to estimate confidence intervals of model parameters. As a consequence, it is difficult to quantitatively study natural populations, where the true history is unknown, and to decide whether inferred demographic events are accurate. This is especially true if data are limited, for instance when only one genome is available.

 

In this study, we consider the distances between heterozygous sites in single diploid samples, deriving novel and simple analytical results for their distribution under varying population histories. These theoretical expectations allow us to efficiently fit and compare parametric models of varying complexity. The most probable model output by our method automatically partitions the demographic history into an optimal number of epochs having arbitrary duration, with parameters and confidence intervals estimated jointly. Such analyses reveal fine-scale variation at the individual level within and across populations, additionally allowing us to explore population structure through time. Our new method and analysis enhance our understanding of the extent to which demographic history can be reconstructed from whole genome sequence data in any species.