A Method for Identification of Pancreatic Cancer through Methylation Signatures in Cell-Free DNA
Metadata[+] Show full item record
Pancreatic cancer has high mortality rates in comparison to other cancers due to limited treatment options and challenges in detecting the disease. Early diagnosis is difficult; successful outcomes are directly tied to detection of the cancer before it can spread throughout the body. Evaluation of circulating cell-free DNA (cfDNA), specifically, detection of circulating tumor DNA (ctDNA), is being explored as an approach for non-invasive ‘liquid biopsy’ that can be deployed widely and cost-effectively to screen for early signs of disease. DNA methylation signatures found in cfDNA can serve as a biomarker for detection of cancer. Previous efforts to detect pancreatic cancer using cfDNA showed limited sensitivity for detection. This manuscript describes work to develop a method for early detection of pancreatic cancer in circulating cell-free DNA using publicly available data from The Cancer Genome Atlas (TCGA) and the Gene Expression Omnibus (GEO). The work includes the development of methods for simulation of samples and cell-free DNA data at multiple ctDNA concentrations as well as a pilot implementation and evaluation of a machine learning model for automated classification of cfDNA samples. Chapter 1 introduces background of pancreatic cancer, cfDNA, liquid biopsy, DNA methylation and DNA methylation in cancer as well as a review of previous efforts to develop early detection applications. Chapter 2 describes the identification of DNA methylation markers that distinguish pancreatic tumor from normal pancreas and blood using publicly available data from the TCGA and GEO. In addition, the chapter outlines the development of a machine learning model to classify samples as tumor or normal based on these markers. Chapter 3 gives an overview of the challenges of genetic data simulation and the development of a novel tool, Heisenberg, for simulating DNA methylation data and cfDNA methylation data. This chapter also illustrate the use of Heisenberg to simulate normal blood samples and pancreatic cancer cfDNA samples. Chapter 4 describes the development of a neural network classification model for detection of pancreatic tumor in different concentrations of cfDNA using simulated samples. The chapter also reports the detection performance of the model using different model training strategies.
Table of Contents
Introduction -- Development of a model to discriminate between pancreatic tumor, normal pancreas, and normal blood -- Heisenberg: A tool for augmenting DNA methylome datasets and simulating cell-free tumor DNA signals -- Evaluation of classification model using simulated datasets -- Appendix
Ph.D. (Doctor of Philosophy)