Modelling dependencies in genetic-marker data and its application to haplotype analysis

Abstract

The objective of this thesis is to develop new methods to reconstruct haplotypes from phaseunknown genotypes. The need for new methodologies is motivated by the increasing avail¬ ability of high-resolution marker data for many species. Such markers typically exhibit correlations, a phenomenon known as Linkage Disequilibrium (LD). It is believed that re¬ constructed haplotypes for markers in high LD can be valuable for a variety of application areas in population genetics, including reconstructing population history and identifying genetic disease variantsTraditionally, haplotype reconstruction methods can be categorized according to whether they operate on a single pedigree or a collection of unrelated individuals. The thesis begins with a critical assessment of the limitations of existing methods, and then presents a uni¬ fied statistical framework that can accommodate pedigree data, unrelated individuals and tightly linked markers. The framework makes use of graphical models, where inference entails representing the relevant joint probability distribution as a graph and then using associated algorithms to facilitate computation. The graphical model formalism provides invaluable tools to facilitate model specification, visualization, and inference.Once the unified framework is developed, a broad range of simulation studies are conducted using previously published haplotype data. Important contributions include demonstrating the different ways in which the haplotype frequency distribution can impact the accuracy of both the phase assignments and haplotype frequency estimates; evaluating the effectiveness of using family data to improve accuracy for different frequency profiles; and, assessing the dangers of treating related individuals as unrelated in an association study.