Discovering novel human structural variation from diverse populations and disease patients: an exploration of what human genomics misses by relying on reference-based analyses

Since the completion of the human genome project, the field of genomics has relied on the human reference genome for nearly all analyses. Population genetics, disease association studies, and beyond all begin by comparing an individual’s sequenced genome to the human reference. However, the human reference genome is not only still incomplete, but also not an accurate representation of humanity; it is derived primarily from a single individual, and cannot possibly represent the scope of human diversity. By using this genome as a template, we bias our studies. In this thesis we examine large regions of structural variation between individuals that are often missed by comparing solely to the human reference genome. We use multiple strategies to uncover variation, including performing localized assembly on whole genome sequencing reads not matching the reference genome from 910 individuals of African ancestry, and utilizing new, long-read sequencing technologies in disease patients. We demonstrate that vast amounts of sequence present in human populations, nearly 300 megabases in the case of the African ancestry dataset, are missing from the reference genome, as well as that many non-reference sequences are present in breast cancer and Mendelian disease patients, which could have yet-to-be-discovered disease relevance. We find evidence of novel non-reference sequences which are genic and transcribed in many individuals, which may have functional relevance. Finally we present strategies for integrating the wealth of short-read sequencing data currently available with the limited but growing number of newer, long-read sequenced samples to gain new insights previously inaccessible using short-read data alone