Date of Award

Fall 12-16-2016

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Programs

Genetics, Cell Biology & Anatomy

First Advisor

Dr. Chittibabu(Babu) Guda

Abstract

Fusion genes are those that result from the fusion of two or more genes, and they are typically generated due to the perturbations in the genome structure in cancer cells. In turn, fusion genes can contribute to tumor formation and progression by promoting the expression of an oncogene, deregulation of a tumor-suppressor, or producing much more active abnormal proteins. More importantly, oncogenic fusion genes are specifically expressed in the tumor cells, which provide enormous diagnostic and therapeutic advantages for cancer treatment. With the development of next-generation sequencing (NGS) technology, RNA-Seq becomes increasingly popular for transcriptomic study because of its high sensitivity and the capability of detecting novel transcripts including fusion genes. To date, many fusion gene detection tools have been developed, most of which attempt to find reliable alignment evidence for chimeric transcripts from RNA-Seq data. It is well accepted that the alignment quality of sequencing reads against the reference genome is often limited when significant differences in the genomes exist, which is the case with cancer genomes that contain many genomic perturbations and structural variations. Hence, regions where fusion genes occur in the cancer genome tend to be largely different from those in the reference genome, which prevents the alignment-based fusion gene detection methods from achieving good accuracies.

We developed a tool called ChimeRScope. ChimeRScope, being an alignment-free method, bypasses the sequence alignment step by assessing the gene fingerprint profiles (in the form of k-mers) from RNA-Seq paired-end reads for fusion gene prediction (Chapter Two). We also optimized the data structure and ChimeRScope algorithms, in order to overcome the common limitations (memory-utilization, low accuracies) that are commonly seen in alignment-free methods (Chapter Two). Results on simulated datasets, previously studied cancer RNA-Seq datasets, and experimental validations on in-house datasets have shown that ChimeRScope consistently performed better than other popular alignment-based methods irrespective of the read length and depth of sequencing coverage (Chapter Three). ChimeRScope also generates graphical outputs for illustrations of the fusion patterns. Lastly, we also developed downloadable software for ChimeRScope and implemented an online data analysis server using the Galaxy platform (Chapter Four). ChimeRScope is available at https://github.com/ChimeRScope/ChimeRScope/.

Share

COinS