Thesis - Open Access
Master of Science (MS)
Mathematics and Statistics
Accurate regulatory DNA motif (or motif) identification plays a fundamental role in the elucidation of transcriptional regulatory mechanisms in a cell and can strongly support the regulatory network construction for both prokaryotic and eukaryotic organisms. Next-generation sequencing techniques generate a huge amount of biological data for motif identification. Specifically, Chromatin Immunoprecipitation followed by high throughput DNA sequencing (ChIP-seq) enables researchers to identify motifs on a genome scale. Recently, technological improvements have allowed for DNA structural information to be obtained in a high-throughput manner, which can provide four DNA shape features. The DNA shape has been found as a complementary factor to genomic sequences in terms of transcription factor (TF)-DNA binding specificity prediction based on traditional machine learning models. Recent studies have demonstrated that deep learning (DL), especially the convolutional neural network (CNN), enables identification of motifs from DNA sequence directly. Although numerous algorithms and tools have been proposed and developed in this field, (1) the lack of intuitive and integrative web servers impedes the progress of making effective use of emerging algorithms and tools; (2) DNA shape has not been integrated with DL; and (3) existing DL models still suffer high false positive and false negative issues in motif identification. This thesis focuses on developing an integrated web server for motif identification based on DNA sequences either from users or built-in databases. This web server allows further motif-related analysis and Cytoscape-like network interpretation and visualization. We then proposed a DL framework for both sequence and shape motif identification from ChIP-seq data using a binomial distribution strategy. This framework can accept as input the different combinations of DNA sequence and DNA shape. Finally, we developed a gated convolutional neural network (GCNN) for capturing motif dependencies among long DNA sequences. Results show that our developed web server enables providing comprehensive motif analysis functionalities compared with existing web servers. The DL framework can identify motifs using an optimized threshold and disclose the strong predictive power of DNA shape in TF-DNA binding specificity. The identified sequence and shape motifs can contribute to TF-DNA binding mechanism interpretation. Additionally, GCNN can improve TF-DNA binding specificity prediction than CNN on most of the datasets.
Library of Congress Subject Headings
DNA -- Analysis.
Nucleotide sequence -- Identification.
Includes bibliographical references (pages 32-46)
Number of Pages
South Dakota State University
In Copyright - Non-Commercial Use Permitted
Yang, JInyu, "Development of Computational Techniques for Regulatory DNA Motif Identification Based on Big Biological Data" (2017). Electronic Theses and Dissertations. 2155.