Superior prognosis is associated with earlier staged diagnosis of ovarian can-cer. Patient prognosis is further affected by histology of epithelial ovarian cancer. Accurate methods to distinguish between these histological subtypesis crucial for faster diagnosis, effective treatment choices and both improvedprognosis and treatment response in patients. Subtypes include clear cell, en-dometrioid, mucinous, high grade and low grade serous. This study trainedand validated linear, radial basis function and polynomial support vectormachines, K-nearest neighbours and random forest models to classify patientsamples in four separate cases: (1) clear cell vs endometrioid, (2) clear cell vs non-clear cell, (3) LGSOC vs non-LGSOC, (4) mucinous vs non-mucinous.Models for each case were trained using gene signatures of lengths 11, 15, 18and 25, respectively. Gene signatures were formed by a selection of differen-tially expressed probes representing genes unique to a case’s specific histologyidentified by non-parametric hypothesis tests (BH adjusted, p2).The polynomial support vector machine model for classifying case (1)performed best achieving a test F1 score of 100%. The best performingmodel for classifying case (2) was a random forest model achieving testingF1 score of 90.9%. The best performing model for classifying case (3) was theradial basis function support vector machine model exhibiting a train andtest F1 score of 100%. A linear support vector machine model performed bestfor classifying case (4), displaying a train and test F1 score of 100%. Basedon the proposed gene signatures in this study, it is indicated that mucinousand non-mucinous samples were completely linearly separable. Moreover,gene signatures suggested for LGSOC and mucinous classifications presentpossible diagnostic capabilities for their corresponding EOC subtype. Clearcell and endometrioid EOC can also be differentiated between by using thesuggested gene signature within this study.Further study on the selected genes in each gene signature is suggestedto determine their role in ovarian cancer and specific histological subtypes ofepithelial ovarian cancer.
| Date of Award | 12 Sep 2024 |
|---|
| Original language | English |
|---|
| Supervisor | Ann Smith (Main Supervisor) & William Lee (Co-Supervisor) |
|---|