UC Berkeley

The full genealogical history of a sample of DNA sequences, resulting from both coalescent and recombination events, is described in a data structure called Ancestral Recombination Graph (ARG). The ARG contains full information needed for answering various types of questions in evolutionary genetics, as opposed to extracting low-dimensional summary statis- tics, such as Site Frequency Spectrum (SFS), Identity-by-Descent (IBD), etc. Thus it is ideal to get accurate estimates of ARGs which can be used for comprehensive downstream evolu- tionary analysis. However, the high dimensionality of the ARG and increasing genomics data have imposed great statistical and computational challenges on the the inference of ARG, to make it usable for real data downstream analysis. In recent years significant progress has been made in the inference of ARG, but the inference accuracy in several key aspects have been still unsatisfactory. In this dissertation, I will show my work on benchmarking existing methods, as well as newly-designed efficient, robust and accurate methods for ARG inference, for both ARG topology and branch length. These methods are applied to real data to learn various aspects of the evolutionary history, including trans-species polymorphism, introgression from archaic hominin, demographic history, mutation rates, and so on.