Document Type

Thesis - Open Access

Award Date


Degree Name

Master of Science (MS)

Department / School

Mathematics and Statistics

First Advisor

Qin Ma


The next-generation sequencing technologies can generate large-scale biological data with higher resolution, better accuracy, and lower technical variation than the arraybased counterparts. RNA sequencing (RNA-Seq) can generate genome-scale gene expression data in biological samples at a given moment, facilitating a better understanding of cell functions at genetic and cellular levels. The abundance of gene expression datasets provides an opportunity to identify genes with similar expression patterns across multiple conditions, i.e., co-expression gene modules (CEMs). Genomescale identification of CEMs can be modeled and solved by biclustering, a twodimensional data mining technique that allows clustering of rows and columns in a gene expression matrix, simultaneously. Compared with traditional clustering that targets global patterns, biclustering can predict local patterns. This unique feature makes biclustering very useful when applied to big gene expression data since genes that participate in a cellular process are only active in specific conditions, thus are usually coexpressed under a subset of all conditions. The combination of biclustering and large-scale gene expression data holds promising potential for condition-specific functional pathway/network analysis. However, existing biclustering tools do not have satisfied performance on high-resolution RNA-Seq data, majorly due to the lack of (i) a consideration of high sparsity of RNA-Seq data, especially for scRNA-Seq data, and (ii) an understanding of the underlying transcriptional regulation signals of the observed gene expression values. QUBIC2, a novel biclustering algorithm, is designed for large-scale bulk RNA-Seq and single-cell RNA-seq (scRNA-Seq) data analysis. Critical novelties of the algorithm include (i) used a truncated model to handle the unreliable quantification of genes with low or moderate expression; (ii) adopted the Gaussian mixture distribution and an information-divergency objective function to capture shared transcriptional regulation signals among a set of genes; (iii) utilized a Dual strategy to expand the core biclusters, aiming to save dropouts from the background; and (iv) developed a statistical framework to evaluate the significances of all the identified biclusters. Method validation on comprehensive data sets suggests that QUBIC2 had superior performance in functional modules detection and cell type classification. The applications of temporal and spatial data demonstrated that QUBIC2 could derive meaningful biological information from scRNA-Seq data. Also presented in this dissertation is QUBICR. This R package is characterized by an 82% average improved efficiency compared to the source C code of QUBIC. It provides a set of comprehensive functions to facilitate biclustering-based biological studies, including the discretization of expression data, query-based biclustering, bicluster expanding, biclusters comparison, heatmap visualization of any identified biclusters, and co-expression networks elucidation. In the end, a systematical summary is provided regarding the primary applications of biclustering for biological data and more advanced applications for biomedical data. It will assist researchers to effectively analyze their big data and generate valuable biological knowledge and novel insights with higher efficiency.

Library of Congress Subject Headings

Gene expression -- Data processing.
Data mining.


Includes bibliographical references



Number of Pages



South Dakota State University



Rights Statement

In Copyright