STUDIES ON VARIABILITY IN CANCER GENE EXPRESSION: FROM SINGLE PROTEINS TO POPULATIONS

Abstract

In this dissertation I describe four projects investigating different aspects of the variability of gene expression in human cancers. In the first chapter, we analyze epidemiological incidence rates for autoimmune diseases and cancers across numerous populations and find that sex biases in incidence rates are positively correlated between autoimmune diseases and cancers arising from the same tissue. We find that across these tissues the expression of protein-codingmitochondrial genes is positively correlated with both autoimmune disease and cancer incidence rate sex biases, suggesting a possible direction for further investigation. In the second chapter, I construct a computational pipeline to conduct unbiased searches in large databases for possible events accounting for cancer neopeptides predicted by mass spectrometry. I identify several ribosomal frameshift-derived neopeptides from HLA-peptidomics data and discuss future approaches for further improving the accuracy and flexibility of our approach. In the third chapter, I compare the power of different multivariate Cox proportional hazards survival models based on gene- and below-gene-level expression measures to predict genes whose expression in tumor samples at diagnosis affects subsequent survival of cancer patients. I find that models based on both gene-level expression and isoform-level expression (whether transcript abundance or relative transcript abundance) identify the greatest number of statistically significant genes of interest. Finally, in the fourth chapter I briefly explore how heteroformity and entropy measures can be used to examine differences in mRNA splicing diversity at numerous levels of comparison. I propose some simple visualizations that harness these measures to display patterns in splicing diversity.