Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


Brian Kennedy Thesis 6-29-17.pdf (3.05 MB)

ETD Abstract Container

Abstract Header

Leveraging Multimodal Tumor mRNA Expression Data from Colon Cancer: Prospective Observational Studies for Hypothesis Generating and Predictive Modeling

Kennedy, Brian Michael, Kennedy
http://orcid.org/0000-0003-0708-9274
http://rave.ohiolink.edu/etdc/view?acc_num=osu1498742562364379

Abstract Details

2017, Master of Science, Ohio State University, Public Health.
Colon cancers are second only to lung cancers in the number of cancer deaths in the United States per annum. Common treatment regimens are surgical resection, optionally followed by adjuvant chemotherapy. Successful outcomes are measured by lack of progression to advanced stages at the primary tumor site and the absence of recurrence elsewhere in the body. Stage 3 cancer patients relapse at 50%, while stage 2 cancer patients relapse 25-40%. The most common site of distant metastasis is the liver; about 11% of patients who relapse survive 5 years. Unfortunately, there are no clear means to predict which patients will relapse following treatment. Subtyping of colon cancers is detailed in literature, although no clear translation to predicting clinical outcome has occurred. A 2014 meta-analysis by the Agency for Healthcare Research and Quality showed existing commercial means of predicting relapse provided dubious benefits to patients. This thesis details a method to create a predictive model of relapse in colon cancer patients that is an improvement over existing standards of care using gene expression patterns in specific stages coordinated with histopathological subtypes to be examined in vitro. We conducted a retrospective analysis of mRNA expression in colon cancer patients at the time of treatment, integrating genomic data from microarray and RNA-seq platforms with matching clinical data. The main focus of this research was genes with bimodal gene expression due to the ability of bimodal genes to fall along tumor subtypes with unique biological, clinical, and prognostic characteristics. Our results successfully identified bimodal genes through a novel ensemble testing system that recognizes clusters of gene expression values that decompose a single Gaussian distribution into two component Gaussian distributions. The utility and efficacy of the method was demonstrated with known bimodal gene markers in breast cancer patients as a positive control as well as simulation studies. We then examined changes in bimodal status over stage, and found conserved patterns over multiple independent datasets. We isolated genes that showed patterns of loss and acquisition of bimodality over stage with the hypothesis these genes may be informative in predicting relapse in colon cancer patients or to progression to advanced stages of the disease. The same analysis was conducted with ovarian cancer patients with genes identified computationally then confirmed functionally by in vitro knockdown studies. With the methodology and preliminary results confirmed by known datasets and in vitro models, we generated potential gene targets for further study in colon cancer patients. Additionally, simple tree-based models of relapse using Boolean gene target variables trained on GEO microarray datasets retained good negative predictive values when tested on TCGA RNA-seq data. These models represent both an improvement in predictive power over existing tests and provide starting points for further examination of biomarkers/gene targets in colon tumors.
Kun Huang, PhD (Advisor)
Randall Harris, MD,PhD (Committee Member)
James Chen, MD (Committee Member)
Joanna Groden, PhD (Committee Member)
89 p.
Bioinformatics; Biostatistics; Medicine; Public Health
colon cancer; bimodal distribution; Gaussian; mRNA; RNA-seq; microarray; relapse; metastasis; classification; subtype; MMR; mismatch repair; fast frugal tree; random forest; upset; breast cancer; ovarian cancer

Recommended Citations

Citations

  • Kennedy, Kennedy, B. M. (2017). Leveraging Multimodal Tumor mRNA Expression Data from Colon Cancer: Prospective Observational Studies for Hypothesis Generating and Predictive Modeling [Master's thesis, Ohio State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=osu1498742562364379

    APA Style (7th edition)

  • Kennedy, Kennedy, Brian. Leveraging Multimodal Tumor mRNA Expression Data from Colon Cancer: Prospective Observational Studies for Hypothesis Generating and Predictive Modeling. 2017. Ohio State University, Master's thesis. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=osu1498742562364379.

    MLA Style (8th edition)

  • Kennedy, Kennedy, Brian. "Leveraging Multimodal Tumor mRNA Expression Data from Colon Cancer: Prospective Observational Studies for Hypothesis Generating and Predictive Modeling." Master's thesis, Ohio State University, 2017. http://rave.ohiolink.edu/etdc/view?acc_num=osu1498742562364379

    Chicago Manual of Style (17th edition)

osu1498742562364379
381
© 2017, all rights reserved.
This open access ETD is published by The Ohio State University and OhioLINK.