Mapping the Landscape of Mutation Rate Heterogeneity in the Human Genome: Approaches and Applications

Abstract

All heritable genetic variation is ultimately the result of mutations that have occurred in the past. Understanding the processes which determine the rate and spectra of new mutations is therefore fundamentally important in efforts to characterize the genetic basis of heritable disease, infer the timing and extent of past demographic events (e.g., population expansion, migration), or identify signals of natural selection. This dissertation aims to describe patterns of mutation rate heterogeneity in detail, identify factors contributing to this heterogeneity, and develop methods and tools to harness such knowledge for more effective and efficient analysis of whole-genome sequencing data.

In Chapters 2 and 3, we catalog granular patterns of germline mutation rate heterogeneity throughout the human genome by analyzing extremely rare variants ascertained from large-scale whole-genome sequencing datasets. In Chapter 2, we describe how mutation rates are influenced by local sequence context and various features of the genomic landscape (e.g., histone marks, recombination rate, replication timing), providing detailed insight into the determinants of single-nucleotide mutation rate variation. We show that these estimates reflect genuine patterns of variation among de novo mutations, with broad potential for improving our understanding of the biology of underlying mutation processes and the consequences for human health and evolution. These estimated rates are publicly available at http://mutation.sph.umich.edu/.

In Chapter 3, we introduce a novel statistical model to elucidate the variation in rate and spectra of multinucleotide mutations throughout the genome. We catalog two major classes of multinucleotide mutations: those resulting from error-prone translesion synthesis, and those resulting from repair of double-strand breaks. In addition, we identify specific hotspots for these unique mutation classes and describe the genomic features associated with their spatial variation. We show how these multinucleotide mutation processes, along with sample demography and mutation rate heterogeneity, contribute to the overall patterns of clustered variation throughout the genome, promoting a more holistic approach to interpreting the source of these patterns.

In chapter 4, we develop Helmsman, a computationally efficient software tool to infer mutational signatures in large samples of cancer genomes. By incorporating parallelization routines and efficient programming techniques, Helmsman performs this task up to 300 times faster and with a memory footprint 100 times smaller than existing mutation signature analysis software. Moreover, Helmsman is the only such program capable of directly analyzing arbitrarily large datasets. The Helmsman software can be accessed at https://github.com/carjed/helmsman.

Finally, in Chapter 5, we present a new method for quality control in large-scale whole-genome sequencing datasets, using a combination of dimensionality reduction algorithms and unsupervised anomaly detection techniques. Just as the mutation spectrum can be used to infer the presence of underlying mechanisms, we show that the spectrum of rare variation is a powerful and informative indicator of sample sequencing quality. Analyzing three large-scale datasets, we demonstrate that our method is capable of identifying samples affected by a variety of technical artifacts that would otherwise go undetected by standard ad hoc filtering criteria. We have implemented this method in a software package, Doomsayer, available at https://github.com/carjed/doomsayer.