Date of Award

Spring 2017

Document Type


Degree Name

Doctor of Philosophy (PhD)


Mathematics, Statistics and Computer Science

First Advisor

Bozdag, Serdar

Second Advisor

Merrill, Stephen

Third Advisor

Maadooliat, Mehdi


Cancer genomes accumulate alterations that promote cancer cell proliferation and survival. Structural, genetic and epigenetic alterations that have a selective advantage for tumorigenesis affect key regulatory genes and microRNAs that in turn regulate the expression of many target genes. The goal of this dissertation is to leverage the alteration-rich landscape of cancer genomes to detect key regulatory genes and microRNAs. To this end, we designed a feature selection algorithm to identify DNA methylation signals around a gene that would highly predict its expression. We found that genes whose expression could be predicted by DNA methylation accurately were enriched in Gene Ontology terms related to the regulation of various biological processes. This suggests that genes controlled by DNA methylation are regulatory genes. We also developed two tools that infer relationships between regulatory genes and target genes leveraging structural and epigenetic data. The first tool, ProcessDriver integrates copy number alteration and gene expression datasets to identify copy number cancer driver genes, target genes of these drivers and the disrupted biological processes. Our results showed that driver genes selected by ProcessDriver are enriched in known cancer genes. Using survival analysis, we showed that drivers are linked to new tumor events after initial treatment. The second tool was developed to leverage structural and epigenetic data to infer interactions between regulatory genes and targets on a network-level. Our canonical correlation analysis-based approach utilized the DNA methylation or copy number states of potential regulators and the expression states of potential targets to score regulatory interactions. We then incorporated these regulatory interaction scores as prior knowledge in a dynamic Bayesian framework utilizing time series gene expression data. Our results indicated that the canonical correlation analysis-based scores reflect the true interactions between genes with high accuracy, and the accuracy can be further increased by using the scores as a prior in the dynamic Bayesian framework. Finally, we are developing an algorithm to detect cancer-related microRNAs, associated targets and disrupted biological processes. Our preliminary results suggest that the modules of miRNAs and target genes identified in this approach are enriched in known microRNA-gene interactions.