On the use of kernel approximate Bayesian computation to infer population history.

Abstract

Genetic variation is a product of mutation, recombination and natural selection along with a complex history involving population subdivision, gene flow and changes in population size. Elucidating the evolutionary forces that shape genetic differences among populations is a major objective of evolutionary genetics. Recent advances in high-throughput technology enable genomic data to be obtained from samples at a population-based scale. Further, the growth in computational power has facilitated extensive efforts to develop intensive simulation-based approaches with the aim of analyzing such large-scale data and making inferences about population history. Approximate Bayesian computation (ABC) provides a quantitative way to assess the goodness-of-fit of complex models that are based on previous knowledge and to estimate the parameters of interest that produce the observed data. The practical advantage of ABC is the application of Bayesian inference to any model without the need to derive a likelihood function. ABC has rapidly become popular in ecology and evolutionary studies due to the contribution it has made to improving computational efficiency over the past decade. This review provides a brief overview of the background of ABC, including potential biases in estimation due to the assumptions and approximation involved, followed by an in-depth review of one of the recently developed ABCs, "kernel ABC," with an explanation of how to overcome these biases. Finally, the application of kernel ABC to the inference of demographic history is summarized.