Computational Methods for Inferring Transcriptome Dynamics Public Deposited

Downloadable Content

Download PDF
Last Modified
  • March 21, 2019
  • Welch, Joshua
    • Affiliation: College of Arts and Sciences, Department of Computer Science
  • The sequencing of the human genome paved the way for a new type of medicine, in which a molecular-level, cell-by-cell understanding of the genomic control system informs diagnosis and treatment. A key experimental approach for achieving such understanding is measuring gene expression dynamics across a range of cell types and biological conditions. The raw outputs of these experiments are millions of short DNA sequences, and computational methods are required to draw scientific conclusions from such experimental data. In this dissertation, I present computational methods to address some of the challenges involved in inferring dynamic transcriptome changes. My work focuses two types of challenges: (1) discovering important biological variation within a population of single cells and (2) robustly extracting information from sequencing reads. Three of the methods are designed to identify biologically relevant differences among a heterogenous mixture of cells. SingleSplice uses a statistical model to detect true biological variation in alternative splicing within a population of single cells. SLICER elucidates transcriptome changes during a sequential biological process by positing the process as a nonlinear manifold embedded in high-dimensional gene expression space. MATCHER uses manifold alignment to infer what multiple types of single cell measurements obtained from different individual cells would look like if they were performed simultaneously on the same cell. These methods gave insight into several important biological systems, including embryonic stem cells and cardiac fibroblasts undergoing reprogramming. To enable study of the pseudogene ceRNA effect, I developed a computational method for robustly computing pseudogene expression levels in the presence of high sequence similarity that confounds sequencing read alignment. AppEnD, an algorithm for detecting untemplated additions, allowed the study of transcript modifications during RNA degradation.
Date of publication
Resource type
Rights statement
  • In Copyright
  • McMillan, Leonard
  • Purvis, Jeremy
  • Hartemink, Alexander
  • Marzluff, William
  • Prins, Jan
  • Jones, Corbin
  • Doctor of Philosophy
Degree granting institution
  • University of North Carolina at Chapel Hill
Graduation year
  • 2017

This work has no parents.