Electronic Thesis and Dissertation Repository

Degree

Doctor of Philosophy

Program

Biochemistry

Supervisor

Dr. Peter Rogan

Abstract

Breast cancer (BC) patient management has been transformed over the last two decades due to the development and application of genome-wide technologies. The vast amounts of data generated by these assays, however, create new challenges for accurate and comprehensive analysis and interpretation. This thesis describes novel methods for fluorescence in-situ hybridization (FISH), array comparative genomic hybridization (aCGH), and next generation DNA- and RNA-sequencing, to improve upon current approaches used for these technologies. An ab initio algorithm was implemented to identify genomic intervals of single copy and highly divergent repetitive sequences that were applied to FISH and aCGH probe design. FISH probes with higher resolution than commercially available reagents were developed and validated on metaphase chromosomes. An aCGH microarray was developed that had improved reproducibility compared to the standard Agilent 44K array, which was achieved by placing oligonucleotide probes distant from conserved repetitive sequences.

Splicing mutations are currently underrepresented in genome-wide sequencing analyses, and there are limited methods to validate genome-wide mutation predictions. This thesis describes Veridical, a program developed to statistically validate aberrant splicing caused by a predicted mutation. Splicing mutation analysis was performed on a large subset of BC patients previously analyzed by the Cancer Genome Atlas. This analysis revealed an elevated number of splicing mutations in genes involved in NCAM pathways in basal-like and HER2-enriched lymph node positive tumours. Genome-wide technologies were leveraged further to develop chemosensitivity models that predict BC response to paclitaxel and gemcitabine. A type of machine learning, called support vector machines (SVM), was used to create predictive models from small sets of biologically-relevant genes to drug disposition or resistance. SVM models generated were able to predict sensitivity in two groups of independent patient data.

High variability between individuals requires more accurate and higher resolution genomic data. However the data themselves are insufficient; also needed are more insightful analytical methods to fully exploit these data. This dissertation presents both improvements in data quality and accuracy as well as analytical procedures, with the aim of detecting and interpreting critical genomic abnormalities that are hallmarks of BC subtypes, metastasis and therapy response.

Thesis Supplementary Tables.xlsx (48832 kB)
Supplementary Tables

Share

COinS