Discovering Molecular Patterns with Therapeutic Implications in Large-Cohort Heterogeneous Cross-Cancer Data

Recent advances in high-throughput genomic technologies and high-performance computing have propelled the science of computational genomics into a new era and launched the field of precision medicine. Computational genomics is now an integral part of biomedical research and genomic testing is routinely performed in clinical settings. In the field of cancer informatics, the integration of genomics has led to invaluable insights and discoveries. We study cancers in order to better understand tumorigenesis and disease progression. This understanding can, in turn, inform and guide therapeutic decisions and suggest directions for drug development and repositioning. The ultimate goal of cancer precision medicine is to sequence and analyze every patients tumor in order to provide the most effective and least toxic treatment.Various experimental platforms are available for collection of different perspectives or views of the cell state, which help us characterize and understand molecular signals driving the cell phenotype. We collectively refer to these views as ’omics’ data. While vast amounts of ’omics’ data are being collected from tumor samples at an accelerating rate, few resources exist to aid biologists and clinicians in identifying trends in these data, finding connections within and between cancer subtypes, and matching patients to previously studied patient groups to infer therapeutic implications. In our analysis we also utilize bioinformatics methods that manipulate, transform, and integrated these views to derive new views of the cell. In my doctoral thesis, I present my work developing new tools and methods to aid the scientific community in understand- ing and interpreting cancer biology (Chapter 2). I also present my work applying such methods to contribute to cancer subtype-specific analyses as part of various projects and collaborations during my doctoral work (Chapter 3).Finally, I describe my work and contributions to the field of personalized medicine in pediatric cancer. While similar in some ways to adult cancers, pediatric cancers differ dramatically from their adult counterparts on a molecular level. For ex- ample, pediatric tumors generally have fewer genomic alterations than adult tumors. Further,childhood cancers are rarer than adult cancers and thus more difficult to study due to a lack of sufficiently large patient cohorts. While some clinics now regularly sequence pediatric tumors for bioinformatic analysis, the sequencing of patient genomes in the clinic is only beginning to impact patient care. Most computational methods for detecting differentially expressed genes are designed for analyzing patient cohorts in research settings and are thus unsuitable for interpreting RNA sequencing data from a single patient. Further, analyzing individuals genomic data leads to actionable treatment options in only fifteen percent of all childhood cancer cases. This is because pediatric cancers are often not driven by non-hereditary genomic changes, and any genomic aberrations that do exist may not be targetable by existing drugs. More sophisticated informatics tools and methods are needed in the field of personalized medicine. To this end, I describe my work developing methods for single-patient analyses in pediatric cancer (Chapter 4). While my methods were developed for pediatric cancers, they may also be used to analyze adult tumors