Abstract

The goal of this study was to gain a deeper understanding of the selective sweep models and the statistical and computational methods that disentangle selective sweeps from neutrality. In the Introduction of the thesis I review the literature on the main approaches that have been developed in the last decade to separate selective sweeps from neutral demographic scenarios. Methods on complete and incomplete selective sweeps are reviewed as well as selective sweeps on structured populations. Further, I analyze the effects of past demographic events, especially bottlenecks, on the genealogies of a sample. Finally, I demonstrate that the ineffectiveness of separating selective sweeps from bottlenecks stems from the lack of robust statistics, and most importantly from the similar genealogies that bottlenecks and selective sweeps may generate locally on a recombining chromosome. In the first chapter I introduce a method that combines statistical tests in a machine learning framework, in order to disentangle selective sweeps from neutral demographic scenarios. The approach uses support vectormachines to learn examples from neutral scenarios and scenarios with selection. I demonstrate that the novel approach outperforms previously published approaches for a variety of demographic scenarios. The main reason for the performance difference is the usage of the scenarios with selection, that are not analyzed by classical statistical methods. In the second chapter of the thesis I present an application of the methods on detecting a selective sweep in the African population of D. melanogaster. Demographic history and ascertainment bias schemes have been taken into account. Results pinpoint to the HDAC6 gene as a target of recent positive selection. This study demonstrates the variable threshold approach, which remedies the tendency of some neutrality tests to detect selective sweeps at the edges of the region of interest. In the third chapter I present the results of the analysis of selective sweeps in multi-locus models. I assume that a phenotypic trait evolves under stabilizing or directional selection. In contrast to the classical models of selective sweeps, the evolutionary trajectory of an allele that affects the trait might belong to one of the three categories: it either fixes, disappears or remains polymorphic. Thereafter, I analyze the properties of coalescent trees and neutral polymorphism patterns that are generated from each of the three categories. I show that for the majority of simulated datasets selection cannot be detected unless the trajectory is either fixed or close to fixation.