Text

theses

Despite the major advances in healthcare over the past century, the successful treatment of cancer has remained a significant challenge, and cancers are the second leading cause of death worldwide behind cardiovascular disease. Breast cancer is the most prevalent cancer in women, and an aggressive and difficult to treat breast cancer variant which tends to appear in younger patient populations is Triple Negative Breast Cancer (TNBC). Post-surgical adjuvant radiotherapy is frequently utilized in patients with TNBC, however little is currently known about which patient populations would significantly benefit from this procedure. In this study we take a deep look at 190 TNBC patient samples from the METABRIC (Molecular Taxonomy of Breast Cancer International Consortium) database, sourced through cBioPortal. Our goal was to identify genetic and clinical features that can be used to better understand the effects of adjuvant radiotherapy on post-surgery TNBC patients, and to build a predictive model for the identification of patient populations which would benefit from post-surgery adjuvant radiotherapy, and those which would receive minimal benefit from this procedure. Exploratory data analysis and Kaplan Meier analysis identified statistically significant genes based on Fisher’s Exact Test and PAM50 (Prediction Analysis of Microarray 50) classifications. KNN machine learning algorithm was used to build the predictive models incorporating both clinical and genetic features. All analyses were carried out on RStudio 3.6.0 and cBioPortal’s Onco-Query Language (OCL). The final optimized model was very efficient, with returned parameters as follows: Accuracy – 94.64%, Kappa Statistic – 88.89%, Sensitivity – 100%, Specificity – 87.50% and AUC(ROC) – 93.75%. Four genes and five clinical make up the core features of the model: the genes are AKT1, MAP3K1, MEN1 and SHANK2, while the clinical features are survival months, survival groups, breast surgery, NPI (Nottingham Prognostic Index) and tumor size. Not only does this model have utility in making TNBC treatment decisions, but it demonstrates that a useful predictive model of cancer therapeutic responses can be constructed using a reasonable number of input parameters, even in a highly heterogeneous disease.

ETD_11093

Ph.D.

Includes bibliographical references

doi:10.7282/t3-qc6q-1j57

ETD doctoral

The author owns the copyright to this work.