PhD Proposal: Novel Computational Methods for Identifying mutation signatures, intracellular microbes and HLA alterations in cancer

As of 2017, cancer was the second leading cause of death in the United States behind heart disease, killing ~600,000 Americans per year. While recent advances in cancer treatment have led to drastic increases in survival for a subset of patients, many patients still do not respond to these treatments. Advances in next-generation sequencing have resulted in a deluge of genomic data from cancer patients, which may hold the key to improving treatments. However, specialized computational techniques are necessary to effectively mine this genomic data. In this dissertation, we present multiple computational approaches that address three computational challenges in the field of computational biology for advancing our understanding and treatment of cancer treatment. Given the importance of reproducibility in biomedical research, we provide publicly available workflows for reproducing the results from our computational approaches.The first chapter of this thesis focuses on identifying mutational signatures, which are patterns of mutations thought to represent underlying mutagenic processes, from a matrix of mutations from large tumor cohorts. Mutation signature extraction is computationally challenging because over fifty distinct signatures have been reported even though the number of mutations detected in a given tumor can range from tens to tens of thousands. A further challenge is that some of these signatures are very similar to each other even though they are associated with distinct tumor covariates. To help address this computational challenge, we introduce the Tumor Covariate Signature Model (TCSM), which is the first mutation signature model to incorporate known tumor covariates. We evaluate TCSM focusing on the previously identified mutation signature 3, which is associated with BRCA1/2 inactivation and whose proposed use as a biomarker for PARP inhibitors is complicated by its similarity to mutation signature 5. By leveraging BRCA1/2 inactivation status, TCSM more accurately assign mutations to signature 3 than existing approaches.The second chapter focuses on the tumor microbiome, which consists of bacteria, fungi and other microbes in the tumor microenvironment. One recent discovery is that the sequencing of human tumors can recover a small percentage of non-human reads. One key computational challenge for leveraging these microbial sequences is to distinguish between microbial reads from microbes in the tumor microbiome and contaminating microbial reads that are introduced during sample preparation or sequencing. To help address this challenge, we develop CSI-Microbes (computational identification of Cell type Specific Intracellular Microbes), which mines microbial sequences from single-cell RNA sequencing (scRNA-seq) experiments and compares abundances across cell-types to distinguish cell-type specific intracellular microbes from contaminating microbes. We validated CSI-Microbes using scRNA-seq data of cells intentionally infected with a known intracellular bacterium, Salmonella. Then, we applied CSI-Microbes to scRNA-seq datasets of tumors, identifying bioinformatic evidence for the existence of cell-type specific intracellular microbes in breast cancer and melanoma.The proposed third chapter is a collaboration to characterize resistance to cancer immunotherapy in patients treated by the Surgery Branch at the National Cancer Institute. We will focus on alterations to the human leukocyte antigen (HLA) system because of its role in antigen presentation. The identification of alterations to the HLA system is computationally challenging because the HLA genomic region is highly polymorphic, which requires specialized tools including one that we propose for the identification of allele-specific transcriptional silencing from RNA-seq data.Examining Committee:

Chair: Dr. Max Leiserson Dept rep: Dr. Furong Huang Members: Dr. Eytan Ruppin (co-advisor) Dr. Rob Patro