: Classifying and Generating Repetitive Elements in the Genome Using Deep Learning

Public Deposited
Analytics

Downloadable Content

Download PDF
Request Version for Screen Reader
Last Modified
  • May 14, 2019
Creator
  • Czejdo, Conrad
    • Affiliation: College of Arts and Sciences, Department of Computer Science
Abstract
  • Repetitive elements are sequence patterns in the genome which are duplicated in large quantity. They serve important functions both in genomic preservation and evolution, leading to the need for their fast and accurate classification. The current gold standard for repeat identification can be achieved by establishing correspondences between a well-annotated library of repetitive elements and a given query sequence. However, annotation quality is highly variable across species. Therefore, for genomes whose repeats are poorly annotated, de novo methods must be used. A common approach of de novo methods is to first check the sequence for protein domain conservation. The presence and order of these protein domains are used as features for an expert-crafted rule-based system or an optimized machine learning classifier. Although de novo approaches have achieved modest success, two problems remain. Firstly, they require lengthy consensus sequences which take time to assemble, and may not be representative of the true diversity of repetitive elements in the sample. Secondly, these approaches are heavily reliant on hand-picking a comprehensive set of protein domains, which may need to be constantly adjusted as new repetitive elements are discovered. In this thesis I show that deep learning models are competitive with pattern matching based approaches at the level of a shotgun sequencing strand for de novo classification of repeat elements. I also explore ways of embedding sequences using deep learning models. Finally, I made these tools available through a web-based interface.
Date of publication
Keyword
DOI
Resource type
Advisor
  • mcmillan, Leonard
Degree
  • Bachelor of Science
Academic concentration
  • Computer Science
Honors level
  • Honors
Degree granting institution
  • University of North Carolina at Chapel Hill
Graduation year
  • 2019
Language
  • English
Date uploaded
  • April 27, 2019

Relations

Parents:

This work has no parents.

In Collection:

Items

Thumbnail Title Date Uploaded Visibility Actions
file details HonorsThesisRev (2).pdf April 27, 2019 Public Download