This dissertation proposes tools and analysis of mutational signatures in human cancer and their application to the stratification of patients for drug response. To provide a comprehensive workflow for preprocessing, analysis, and visualization of mutational signatures, I created the Mutational Signature Comprehensive Analysis Toolkit (musicatk) package. musicatk enables users to select different schemas for counting mutation types and easily combine count tables from different schemas. Multiple distinct methods are available to deconvolute signatures and exposures or to predict exposures in individual samples given a pre-existing set of signatures. Additional exploratory features include the ability to compare signatures to the COSMIC database, embed tumors in two dimensions with UMAP, cluster tumors into subgroups based on exposure frequencies, identify differentially active exposures between tumor subgroups, and plot exposure distributions across user-defined annotations such as tumor type. I then use musicatk to analyze the largest tumor sequencing dataset from a Chinese population to date. I identified differences in the levels of signature exposures compared to similar data from a Western cohort. Specifically, COSMIC signature SBS25 was higher in the Chinese dataset for Melanoma and Renal Cell Carcinoma patients and Melanoma patients had lower levels of SBS7a/b (Ultraviolet Light). My analysis also revealed a putative novel signature enriched in pancreatic cancers. Lastly, I assess the ability of mutational signatures to identify patients who may respond to irofulven, a drug for late-stage cancer patients who have defects in the Transcription Coupled Nucleotide Excision Repair (TC-NER) pathway. As the functional understanding of which mutations successfully disrupt this pathway is incomplete, I develop an approach that classifies patients based on evidence of this pathway being disrupted based on levels of mutational signatures. I build a model that successfully predicts patients who will respond to treatment without a known relevant mutation in the TC-NER pathway. The work from this study furthers our understanding of mutational signatures in different populations and demonstrates the feasibility of using mutational signatures to identify patients eligible for drug trials.

Cancer may be treated with personalized medicine, meaning that specific patientsmight respond better to specific treatments instead of having a common treatment. TheReference Drug-based Neural Network (RefDNN) predicts whether a particular cancer cellline will resist a determined drug, but it fails to provide an explanations for this prediction.The thesis objective is to research on explainable machine learning methods to extractrule-based explanations from the RefDNN predictions and conclude on how confident wecan be about these explanations and whether they make sense from a biological point ofview. One of such explainable machine learning methods is Local Rule-based Explanation(LORE), which extracts rule-based explanations from any black box model using localdecision trees. In this thesis LORE is applied to explain the predictions of the RefDNNon a drug sensitivity dataset and three experiments are set up. First experiment tests theaccuracy and general performance of the extracted rule-based explanations. Second experimentstests the robustness of the rule-based explanations. Third experiments checks theglobal fidelity of the local decision trees used by LORE to mimic the RefDNN behaviour.Finally, one rule-based decision is explained from a biological point of view and conclusionsare made on the obtained results.

<p>  </p> <p>Epstein-Barr virus (EBV) is a complex oncogenic symbiont. The molecular mechanisms governing EBV carcinogenesis remain elusive and the functional interactions between virus and host cells are incompletely defined. Some of the known mechanisms include viral integration into the host genome, expression and mutation(s) of viral genes and the host response to the virus. Despite decades of research there is a lack of effective treatment options for EBV-positive cancer patients underscoring an urgent need to further investigate the mechanisms underlying tumorigenesis as well as explore and develop personalized treatment strategies for patients with EBV-positive cancers. In Chapter 1, I introduce Epstein-Barr Virus (EBV), the two phases of EBV lifecycle and an overview of certain EBV-associated carcinomas. I will also discuss the underlying mechanisms and few current therapeutic strategies against EBV infection. Next, I will discuss some of the preclinical model systems and high-throughput computation techniques that are commonly used by researchers in the field of EBV.  </p> <p>In chapter 2, we have systematically analyzed RNA-sequencing from >1000 patients with 15 different cancer types, comparing virus and host factors of EBV+ to EBV- tissues to reveal novel insights into EBV-positive tumors. First, we observed that EBV preferentially integrates at highly accessible regions of the cancer genome with significant enrichment in super-enhancer architecture. Second, we determined that the expression of twelve EBV transcripts, including LMP1 and LMP2, correlated inversely with EBV reactivation signature. Over-expression of these genes significantly suppressed viral reactivation, consistent with a ‘Virostatic’ function. Third, we identified hundreds of novel frequent missense and nonsense variations in Virostatic genes in cancer samples, and that the variant genes failed to regulate their viral and cellular targets in cancer. Lastly, we were able to dichotomously classify EBV-positive tumors based on patterns of host interferon signature genes and immune checkpoint markers, such as PD-L1 and IDO1. </p> <p>In chapter 3, we probed the lifecycle of EBV on a cell-by-cell basis using single cell RNA sequencing (scRNA-seq) data from six EBV-immortalized lymphoblastoid cell lines (LCL). While the majority of LCLs comprised cells containing EBV in the latent phase of its life cycle, we identified two additional clusters that had distinct expression of both host and viral genes. Both clusters were high expressors of EBV Latent Membrane Protein-1 (LMP1) but differed in their expression of other EBV lytic genes, including glycoprotein gene GP350. We further probed into the transcriptional landscape of these clusters to identify potential regulators which will be discussed in further detail in the chapter. Importantly, I was able to demonstrate enhancing HIF1-a signaling by using Pevonedistat, a compound that stabilized HIF1-a can preferentially induce the transcriptional program specific to one of the three identified clusters. </p> <p>In Chapter 4, I describe some of my recent work. In this project, we have used an intuitive <em>in-silico </em>drug prediction approach to rapidly screen and identify FDA-approved or clinically available compounds that can be repurposed to induce lytic cycle in different EBV+ tumors. Using this strategy, we identified Ciclopirox, an antifungal drug, as a potent inducer of lytic cycle in EBV+ epithelial cancers. We used EBV+ GC cells to determine the effect of Ciclopirox on EBV reactivation as well as identify the underlying mechanisms. In summary, we discovered that reactivation of EBV lytic cycle by Ciclopirox is mediated by multiple pathways, two of the major ones being the HIF1-a and NF-kB pathways. Although, Ciclopirox treatment enhanced the killing effect of antiviral, further investigation is needed to effectively deliver this drug <em>in vivo.</em> Throughout this chapter, I have discussed findings that needs further investigation and proposed necessary experiments. Finally, in Chapter 5 I have summarized my work and described how our work can provide novel insights that can help delineate some of the complexities of host-pathogen interactions in EBV-associated malignancies. </p>