Using Rare Genetic Variation to Understand Human Demography and the Etiology of Complex Traits.

Abstract

Modern sequencing technology has revolutionized almost every aspect of human genetics research. Among the novel findings made possible by the sequencing of large samples is how abundant extremely rare genetic variation is in the human genome. Rare genetic variants are likely to have arisen recently. Thus, they provide novel information about recent population history, and because selection has had little time to act on them, sets of rare variants are potentially enriched with important regulatory and biologically functional variants. Detecting associations between rare variants and genetic traits is challenging; conventional single marker association statistics have little power at low allele counts. Several statistics that aggregate information from multiple variants to increase power and detect group-wise associations have been proposed. In chapter 2 we address the robustness of these group-based tests to population stratification. Using the joint site frequency spectrum of samples from multiple European populations, we show that group-based tests cluster into two classes, and p-value inflation in each class is correlated with a specific form of population structure. An abundance of rare genetic variation is evidence of recent population growth. Large sequencing studies have found the frequency spectra they observe in their samples are inconsistent with models of simple exponential growth, likely due to a recent acceleration in the growth rate. To address this, in chapter 3 we propose a two-parameter model of accelerating, faster-than-exponential population growth and incorporate it into the coalescent. We show that our model can generate samples containing large quantities of rare genetic variants without inflating the quantity of more common variants, making them well suited to modeling the recent history of humans. In chapter 4 we develop a series of analytic calculations that allow us to directly sample internal and external branches from a sample's genealogy without resorting to full coalescent simulations. We show that for constant size populations an exact probability function can be defined for branch lengths, and that by using the expected times between coalescent events we can expand our method to a broader range of demographic models.