Abstract
Disentangling the effects of selection and drift is a long-standing problem in population genetics. Simulations show that pervasive selection may bias the inference of demography. Ideally, models for the inference of demography and selection should account for the interaction between these two forces. With simulation-based likelihood-free methods such as Approximate Bayesian Computation (ABC), demography and selection parameters can be jointly estimated. We propose to use the ABC-Random Forests framework to jointly infer demographic and selection parameters from temporal population genomic data (e.g. experimental evolution, monitored populations, ancient DNA). Our framework allowed the separation of demography (census size, N) from the genetic drift (effective population size, N e) and the estimation of genome-wide parameters of selection. Selection parameters informed us about the adaptive potential of a population (the scaled mutation rate of beneficial mutations, θ b), the realized adaptation (the number of mutations under strong selection), and population fitness (genetic load). We applied this approach to a dataset of feral populations of honey bees (Apis mellifera) collected in California, and we estimated parameters consistent with the biology and the recent history of this species.