ALAGBE, Emmanuel Oluwatoba and Covenant University, Theses Masters (2023) A MULTI-OMICS CLASSIFIER FOR PREDICTION OF ANDROGEN DEPRIVATION TREATMENT RESPONSE IN PROSTATE CANCER DATASE. Masters thesis, Covenant University.
![[img]](http://eprints.covenantuniversity.edu.ng/style/images/fileicons/application_pdf.png) |
PDF
Download (539kB) |
Abstract
Prostate cancer (PCa) is estimated to cause over 375,000 deaths and nearly 1.4 million new cases globally. Several factors contribute to PCa heterogeneity, consequently, the stage of the disease decides the strategy employed in combating the disease. The problem of missing data frequently plagues clinical research. The primary treatment outcomes in the TCGA prostate cancer phenotypic dataset had 120 (19.26%) missing values. Treatment strategies could be negatively impacted by limited care giver experience, “trial and error” approaches to treatment, and the genetic makeup of an individual. To the best of our knowledge, using Machine Learning (ML) to forecast treatment response among PCa patients had not been investigated. The aim of this study is to develop a classifier (that acts as a decision support system) from multi-omics datasets for predicting treatment response in PCa patients. RNAseq, miRNAseq, reverse phase protein array (RPPA), copy number variation (CNV) were used in the study. This study employed R programming to preprocess the data. Differential expression analysis for the RNAseq and miRNAseq conducted using the DESeq2 library. Python programming was used to implement the ML algorithms which include XGBoost, Adaboost, multilayer perceptron, decision tree, logistic regression, support vector machine, gradient boosting classifier, Random forests, naive bayes, and K -nearest neighbors. The performance metrics used include macro f1 score, macro recall, macro precision, weighted f1 score, weighted recall, weighted precision, specificity, sensitivity, accuracy, and area under the receiver operator curve. It was discovered that tree-based models were better for the task than probability and kernel-based models. This study computationally demonstrated that muti-omics strategies are generally superior to single-omics strategies, but the adoption of such strategy isn’t a foolproof solution. A classifier capable of predicting treatment outcomes amongst PCa patients was built and the predicted labels for patients with missing phenotypic values in the TCGA dataset was provided.
| Item Type: | Thesis (Masters) |
|---|---|
| Uncontrolled Keywords: | Prostate cancer, Precision Oncology, Multi-omics, Machine Learning Treatment response. |
| Subjects: | Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA76 Computer software R Medicine > R Medicine (General) |
| Divisions: | Faculty of Engineering, Science and Mathematics > School of Electronics and Computer Science |
| Depositing User: | nwokealisi |
| Date Deposited: | 25 Sep 2024 13:04 |
| Last Modified: | 25 Sep 2024 13:04 |
| URI: | http://eprints.covenantuniversity.edu.ng/id/eprint/18454 |
Actions (login required)
 |
View Item |
