Joint likelihood-free inference of the number of selected SNPs and the selection coefficient in an evolving population

Abstract

Given the often intractable exact likelihood, likelihood-free inference plays an important role in population genetics. Indeed, several methodological developments in approximate Bayesian Computation (ABC) were inspired by applications in population genetics. Here, we explore a novel combination of recently proposed ABC tools that can handle high-dimensional summary statistics and apply them to infer selection strength and the number of selected loci from experimental evolution data. While several methods infer selection strength at the single-nucleotide polymorphism (SNP) level, our approach provides additional information about the selective architecture, including the number of selected positions in a candidate window of interest. This is nontrivial, since the spatial correlation induced by genomic linkage can produce selection signals at neighboring SNPs. A further advantage of our approach is that we can readily quantify uncertainty using the ABC posterior. On both simulated and real data, we demonstrate promising performance. This suggests that our ABC variant could also be interesting in other applications.