Inferring Bottlenecks from Genome-Wide Samples of Short Sequence Blocks

The advent of the genomic era has necessitated the development of methods capable of analyzing large volumes of genomic data efficiently. Being able to reliably identify bottlenecks--extreme population size changes of short duration--not only is interesting in the context of speciation and extinction but also matters (as a null model) when inferring selection. Bottlenecks can be detected in polymorphism data via their distorting effect on the shape of the underlying genealogy. Here, we use the generating function of genealogies to derive the probability of mutational configurations in short sequence blocks under a simple bottleneck model. Given a large number of nonrecombining blocks, we can compute maximum-likelihood estimates of the time and strength of the bottleneck. Our method relies on a simple summary of the joint distribution of polymorphic sites. We extend the site frequency spectrum by counting mutations in frequency classes in short sequence blocks. Using linkage information over short distances in this way gives greater power to detect bottlenecks than the site frequency spectrum and potentially opens up a wide range of demographic histories to blockwise inference. Finally, we apply our method to genomic data from a species of pig (Sus cebifrons) endemic to islands in the center and west of the Philippines to estimate whether a bottleneck occurred upon island colonization and compare our scheme to Li and Durbin's pairwise sequentially Markovian coalescent (PSMC) both for the pig data and using simulations.