In the fourteen years since the announcement of the rough draft of the human genome there has been a precipitous drop in the cost of genome sequencing, from several billions of dollars down to a few thousand. This has lead to a new age of data prosperity in biological sciences and to a large growth in the field of Bioinformatics. The extremely low cost of sequencing means that multitudes of species can now be sequenced and those data can be added to the ever-expanding library. This sudden wealth of raw data at a low level in the analysis hierarchy has given an urgency to important methodological questions. What is the best method to assemble a new genome? Given that you have several assembled genomes and you want to know how they're related to one another, what is the best way to align these genomes? And finally, once in possession of an alignment of genomes, how can we best use that information to annotate genes and other biologically interesting regions upon those genomes?

The work presented herein is an attempt to quantify and compare the methods for carrying out these fundamental research tasks. In the first chapter (on the Assem- blathon, a collaborative competition that brought together the world's genome assembly expects), I investigate the problem of short read whole genome de novo assembly. In the second chapter (on the Alignathon, a collaborative competition that brought together the world's whole genome alignment experts) I investigate the problem of whole genome alignment. Finally I address the problem of comparative annotation by using a dataset comprised of the genomes of 17 strains of inbred mouse and the rat in order to establish a quick method of annotating multiple genomes using a combination of evolutionary homology information, existing annotations and extrinsic data (e.g. RNAseq) using off the shelf software components for the majority of the work.

Much of our current understanding of cancer has come from investigating how normal cells are transformed into malignant cancers through the stepwise acquisition of somatic genomic abnormalities. These abnormalities include single nucleotide variants (SNVs), insertions and deletions (INDELs), chromosomal rearrangements, and copy number aberrations. The detection of SNVs is a crucial component to the characterization of the cancer genome. They assist in identifying key genes as possible drug targets, diagnostic markers for early detection, and prognostic markers for monitoring a patient's response to therapy. Variant calling algorithms thus far have focused on comparing the normal and tumor genomes from the same individual. In recent years, it has become routine for projects like The Cancer Genome Atlas (TCGA) to also sequence the tumor RNA. A novel computational method called RADIA (RNA and DNA Integrated Analysis) that combines the patient-matched normal and tumor DNA with the tumor RNA to detect SNVs is presented here. RADIA has detected somatic mutations for nearly 4,500 patients across 22 different cancers, and including the RNA provided a 2-7% increase in sensitivity.

RNA editing is an additional epigenetic mechanism involved in cancer development and progression. RNA editing of the AZIN1 gene has been identified as a driver in the pathogenesis of hepatocellular carcinoma and may be a potential driver for other human cancers as well. An investigation of AZIN1 RNA editing in data collected from nearly 5,000 patients across 12 cancers has been performed. Higher editing frequencies significantly correlated with clinical data such as larger tumor sizes, greater lymph node involvement, the presence of metastases, and higher tumor grades. They were also associated with subtypes that often have the worst prognosis. Over-editing in many cancers is correlated with poor overall and recurrence free survival.

With projects like TCGA providing sequencing data for both DNA and RNA from the same patients across multiple cancers, it is now possible to characterize germline variants, somatic mutations, and RNA editing events on a genome-wide scale. The identification of SNVs that occur in specific genes across multiple cancers provides a powerful way to discover genes that are important to these diseases.

Major Histocompatibility Complex (MHC) class I molecules present intracellular peptides for surveillance by cytotoxic T cells. To fully understand the mechanism of T cell recognition and prioritize different therapeutic targets, it is of practical relevance to first characterize the surface features of peptide/MHC (pMHC) complexes which requires high-resolution structural models. In this work, I present RosettaMHC, a high-throughput pMHC homology modeling approach, which utilizes structural templates from the Protein Data Bank (PDB) selected using several strategies. The accuracy of a peptide backbone is primarily dependent on the selection of a closest template. Here, we have explored template selection based on (i) the peptide sequence, and (ii) the MHC groove sequence which allow us to obtain models within 1.5 Å and 1 Å respectively from their native X-ray structures. In addition, we have added Artificial Neural Network (ANN) models to filter structures that are potentially inaccurate thereby increasing throughput while helping us select models within 1 Å from their corresponding natives. We applied RosettaMHC to model structures of (i) anaplastic lymphoma kinase neoepitopes found in neuroblastoma cancer patients in complex with 2,904 HLA alleles to identify putative alleles that may display these peptides, and (ii) SARS-CoV-2 peptide/HLA-A*02:01 complexes that can be utilized as candidates to generate pMHC tetramers to probe T cells and possibly as vaccine targets. Through this work, we want to establish that, for the most frequent alleles, we have sufficient knowledge from the existing databases to predict sub-angstrom accuracy models thereby reducing the efforts required to experimentally determine pMHC structures. Alongside RosettaMHC, I (i) helped develop resonance assignment methods that are used to interpret NMR data, (ii) worked with alternative protocols in Rosetta to solve structures of biologically important protein targets of varying sizes and complexities with the help of sparse NMR data, and (iii) built a method to estimate side chain conformational entropy using boolean satisfiability and applied it to understand dynamics of the therapeutic cytokine, Interleukin-2, which helped characterize a minor conformational state that is relevant for drug design. Although the Rosetta energy function together with ANN-based scores help us eliminate inaccurate peptide backbones, poor templates may still be locally optimal, leading to imprecise modeling results in some cases. In the future, we want to employ our satisfiability-based side chain packing method towards a more accurate peptide backbone identification. I believe that this work can provide a rational approach for high-throughput modeling of pMHC targets with desired recognition features which are relevant for cancer immunotherapy and infectious diseases.

Comparing related (homologous) subsequences between genomes from different species gives insight into their function. This information is captured in ``genome alignments'', which are essential for almost all comparative genomics analyses. However, most existing methods to create a genome alignment suffer from reference-bias (where only one genome is fully aligned to all others), or ignore duplication events. Though the Cactus genome aligner avoided these restrictions, it could not align more than a few genomes without becoming cost-prohibitive as well as losing accuracy. I developed and refined a “progressive alignment” extension to Cactus to allow it to produce a full alignment in time linear in the number of input genomes while maintaining similar, or often improved, quality. This new method allows Cactus to align hundreds of large vertebrate genomes---enabling comparative genomics at an unprecedented scale. During its development I used Cactus as an essential component of several successful comparative genomics projects. Working closely with the 200 Mammals and Bird 10K projects, I have used Cactus to create an alignment of over 600 bird and mammal genomes, which is by far the largest genome alignment ever created. Finally, I have utilized this alignment to provide a highest-possible-resolution annotation of mammalian and avian evolutionary constraint, using the uniquely large number of taxa to enable the examination of weak effects of purifying selection.

Cancer is a deadly disease that affects 1 in 2 men and 1 in 3 women in the United States. The development of high throughput sequencing technologies has allowed for great advancements in cancer therapy, though, we still have a long way to go. Many cancer patients do not receive care soon enough due to late diagnosis, they do not receive the optimum treatment for their specific tumor type or they receive the optimum treatment, but then develop resistance to it. This thesis will address these three problems and describe how high throughput sequencing technology has allowed us to continue to make advancements in cancer treatment. Specifically, I will describe one example of how understanding the mechanism of oncogenesis can lead to new options for early diagnosis and monitoring of disease progression in about a quarter of glioblastoma multiforme and IDH wild-type lower grade glioma patients, harboring double minute chromosomes. I show how comprehensive characterization of lower grade glioma patients has lead to the discovery of a new, more accurate method of subtyping patients, which will allow for patients to receive more appropriate treatment for their specific tumor type. This is especially important for one subtype of lower grade glioma patients who have poor prognosis and appear very similar to glioblastoma multiforme patients, and therefore should receive similar treatment. I will also describe how development of HER2+ breast cancer cell lines that are resistant to the targeted therapy lapatinib, has led to greater understanding of a mutation-independent mechanism of drug resistance involving activation of the Notch pathway. Patients with a similar mechanism of resistance may benefit from treatment with Notch inhibitors, several of which are currently being tested in clinical trials. All of these findings display ways in which high throughput sequencing can be used to make advances in our understanding of cancer, ultimately moving towards better treatment for patients and better patient outcomes.

Pediatric cancers are different from adult cancers in that they often have few targetable DNA mutations, and in most cases are thought to be developmental in origin rather than environmental. While overall survival rates for pediatric cancer have increased in the past few decades, there remain several difficult-to-treat pediatric cancer types, including deadly pediatric diffuse midline and brainstem gliomas. However, advances in genomics data generation and analysis methods have made it possible to start identifying the dominant signaling pathways, developmental origins, and ultimately therapeutic vulnerabilities of these tumors.

Here I present my work in the UCSC Treehouse Childhood Cancer Initiative using comparative gene expression analysis methods to identify a rare cancer subtype with treatment and outcome implications (Chapter 2). This work demonstrates the utility of gene expression data for molecular subtyping in the clinic, especially in rare or difficult-to-diagnose pediatric cancers. I also present my work identifying a developmental window of opportunity for the histone H3 K27M mutation event, which characterizes a majority of brainstem gliomas and is associated with especially poor overall survival (Chapter 3). The cell-of-origin and developmental timing for gliomagenesis in this tumor type has been a subject of much research in the past decade, and my work helps to address this clinically relevant issue.

Finally, in Chapter 4 I present an analysis of multiple types of genomic data across thousands of primary tumors and diverse cancer laboratory models (cell lines, organoids, and patient-derived mouse xenografts). I report that cancer organoids have the advantage of being relatively accurate cellular representations of primary tumors, and time- and cost-effective. Therefore, in Chapter 5 I present the development of a novel human embryonic stem cell based cerebral organoid model of the histone H3 K27M mutation in early human embryonic brain development. This model can be used to answer outstanding questions in the field to better help treat these devastating tumors.

Advancements in technologies have resulted in an explosion of data, the volume of which continues to increase at an exponential rate. The accumulating wealth of data is enabling numerous new research possibilities and is transforming the world profoundly. In genomics, new genomes are being regularly sequenced, with a growing number of individual genomes becoming available for many species. As the ability to have complete genomic information becomes the norm, the need for a reference genome that better represents the particular species population intensifies: It becomes important to utilize the newly emerged sequences to improve current references and ensure better quality for future assemblies and experiments. Additionally, the proliferation of data has necessitated the decentralization of computational resources together with the empowerment of users to a do-it-yourself system, in which users create their own assemblies, alignments, visualizations and analyses. This is because with the accelerating amount of data, it is impossible and undesirable to maintain the infrastructure model in which only a number of specialized institutions handle most if not all of the data and analyses.

Joining many other on-going efforts, the works in this dissertation attempt to address some of these rising demands. First, I describe the problem of constructing a pan-genome reference for a population and demonstrate that the resulting pan-genome reference is more representative of the population than is any individual genome, using both simulated and real data. Second, I describe a comparative genomic framework that allows for easy generation of collections of web accessible UCSC genome browsers interrelated by an alignment. The pipeline, named the comparative assembly hub (CAH) pipeline, is intended to democratize UCSC comparative genomic resources and facilitate public sharing via the internet. As a demonstration, I create comparative assembly hubs for 66 Escherichia coli/Shigella genomes and highlight comparative analyses on their pan-genomic, core genomic and phylogenetic relationships. Last, I report on comprehensive assessments of the T cell receptor (TCR) repertoires of the autoimmune disease Ankylosing Spondylitis and show example comparative analyses for finding evidence of antigen selection and identifying potential disease-associated clones. In addition, I describe an open-source software package for profiling and comparing TCR sequencing data, called the “Adaptive IMmunoSequencing ToolKit”, or the aimseqtk package. The aimseqtk package is comprised of four main components addressing common analyses of this type of data: clone tracking, repertoire profiling, public clone identification and publication mining.

The recent introductions of low-cost, long-read and read-cloud sequencing tech- nologies coupled with intense efforts to develop efficient algorithms have made affordable, high-quality de-novo sequence assembly a realistic proposition. The result is an explo- sion of new, ultra contiguous genome assemblies. To compare these genomes we need robust methods for genome annotation. I describe the fully open source Comparative Annotation Toolkit (CAT), which provides a flexible way to simultaneously annotate entire clades and identify orthology relationships. I show that CAT can be used to improve annotations on the rat genome, annotate the great apes, annotate a diverse set of mammals, and annotate personal, diploid human genomes. I demonstrate the resulting discovery of novel genes, isoforms and structural variants, even in genomes as well studied as the rat and great apes, and how these annotations improve cross-species RNA expression experiments.

The human immune system can identify malfunctioning, damaged, or infected cells within the body. The integrity of nucleated cells is communicated to the immune system through short peptides which they display on their exterior using MHC proteins. Mutated peptides produced in cancer cells due to single nucleotide events, short insertion or deletion events, or due to large chromosomal rearrangements can help the immune system identify tumor cells as threats. Adoptive T-cell and Vaccine-based therapies that stimulate autologous T-cells to treat a patient's disease are more specific than standard chemo- and radiotherapies because they directly target tumor-specific mutations. They also potentially protect the body from recurrence of the tumor due to their unique memory potential. However, T-cells are often perceptive to only those cells originating from the patient's body, making this a personalized treatment for the disease. Rapid identification of tumor neoepitopes in a patient diagnosed with cancer will greatly reduce the time taken to develop a personalized therapy.

This thesis primarily covers ProTECT, a computational workflow for the Prediction of Epitopes for Cancer Therapy. ProTECT is an automated, scalable, and reproducible end-to-end pipeline that processes patient sequencing data to produce a ranked list of neoepitopes of therapeutic significance. ProTECT also attempts to predict the state of the MHC presentation pathway in the tumor, and the tumor response to immunotherapies based on previously published immune signatures. I describe ProTECT and demonstrate its features using a cohort of 326 Prostate Adenocarcinoma patients. I then demonstrate its clinical utility through the identification of a neoepitope arising from a hotspot mutant in pediatric Neuroblastoma.

To enable ProTECT, I worked on two auxiliary projects, Toil and TransGene. Toil is a workflow manager developed at the Genomics Institute at UCSC, and is the backbone of the ProTECT workflow. I spent some of my time at UCSC working on enhancing the efficiency of Toil workflows. TransGene is a tool I wrote that translates genomic variants (single nucleotide, short insertions and deletions, and fusion genes) into peptides that are compatible with existing peptide:MHC binding affinity predictors. These methods are also included in this thesis.

The scale of the problems which human genomics is asked to solve necessitates that the field develop an ability to integrate and synthesize information across the entire human population. The abstraction of a single-copy human reference genome assembly, and the linear coordinate space that it induces, are more of a hindrance than a help at these scales. They can only ever represent one sample at any given place, and they make combining information about human variation across multiple studies and modalities difficult. To rectify these problems, I propose the construction and adoption of a graph-based alternative to the human reference genome assembly: a Human Genome Variation Map. I present here four research projects. The first is a theory of mapping to references that is extensible to graphs. The second describes a novel data structure for embedding individual haplotype sequences into a graph reference. The third surveys graph construction techniques to discover methods that produce graphs yielding read mapping and variant calling results superior to those obtained with linear, variation-free references. The fourth extends these improvement results to chromosome-scale graphs constructed from multiple sources and modalities of variation data. These four projects describe a research program aimed towards the eventual release of an official Human Genome Variation Map build, providing a piece of vital infrastructure for the analysis of human genomic variation at population scale.