University of Ioannina
Multidimensional computational methods for modeling cancer diagnosis, prognosis and treatment
Abstract
dc:descriptionThe present thesis deals with the modeling of cancer diagnosis, prognosis and treatment by utilizing and implementing well-established computational approaches that can efficiently and effectively contribute to cancer care research and precision oncology. The main objective of this thesis is to study and further understand the molecular basis underlying cancer progression and risk prediction by combining high-throughput data with patient information. Towards this direction, we seek to investigate how the integration of heterogeneous datasets related to cancer development, such as genomic changes and single nucleotide polymorphisms, could provide subsequently a better understanding on cancer classification and progression based on Dynamic Bayesian Networks (DBNs) and ensemble Machine Learning (ML) methodologies, respectively. The first part of the thesis concerns the interactions of the molecules and especially of differentially expressed genes (DEGs) that contribute to cancer progression. Based on this knowledge the identification of DEGS and their related molecular pathways is therefore of great importance. We exploited DEGs in order to further perform pathway enrichment analysis. According to our results we found significant pathways in which the disease associated genes have been identified as strongly enriched. Based on the performed pathway analysis we further proposed a methodology for predicting oral cancer recurrence using DBNs. The methodology takes into consideration time series gene expression data in order to predict a disease recurrence. Subsequently, we can conjecture about the causal interactions between genes in consecutive time intervals. A considerable overall performance of the predictive models was achieved with reference to the knowledge obtained from the pathway level. Cancer classification through DBN-based approaches that could reveal the importance of exploiting knowledge from statistically significant genes and key regulatory molecules was also explored. We identified the genes that act as regulators and mediate the activity of transcription factors that have been found in all promoters of our list with DEGs. These features serve as potential priors for distinguishing tumour from normal samples using a DBN-based classification approach. We employed three microarray datasets from the Gene Expression Omnibus (GEO) public functional repository and performed differential expression analysis. Promoter and pathway analysis of the identified genes revealed the key regulators which influence the transcription mechanisms of these genes. We applied the DBN algorithm on selected genes and identified the features that can accurately classify the samples into wild type and controls. Both accuracy and area under the receiver operating characteristic (ROC) curve (AUC) were high for the gene sets comprising of the DEGS along with their master regulators. In the second part of the thesis we explored the contribution of the genetic susceptibility patients’ profiles and by combining them with known clinical, histological and serological risk factors we enhanced the accuracy of predicting lymphoma development in this patient population. The potential predictive role of both genetic variants and laboratory risk factors were investigated through a ML-based framework which encapsulated ensemble classifiers, such as Gradient Boosting (GB) and Random Forests (RFs) with Gini and entropy measures. Ensemble methods enhance the classification accuracy with approaches that are sensitive to minor perturbations in the training phase. The evaluation of the proposed methodology based on a 10-fold stratified cross validation procedure yielded considerable results in terms of balanced accuracy. The initial clinical, histological and serological findings at an early diagnosis were exploited to establish ML-based predictive tools in clinical practice and further enhance our understanding towards cancer development. In the present thesis, we studied the potential of integrating transcriptomic data with knowledge from the pathway level to model cancer progression and patient risk stratification. The development and application of novel DBN-based analysis methods allowed to infer models that could classify different phenotypes into groups with high classification accuracy. We also demonstrated that robust ensemble ML-based models could contribute to the prediction of cancer development based on the integration of genotype data along with clinical information; thus, contributing to improved disease prognosis and treatment.
Degree
thesis:*- Grantor dc:publisher
- University of Ioannina
- Year dc:date
- 2020
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Kourou, Konstantina
- Κούρου, Κωνσταντίνα
Subjects
dc:subject × 24- Υπολογιστικές μέθοδοι
- Μοντελοποίηση εξέλιξης καρκίνου
- Διαφορικά εκφρασμένα γονίδια
- Ανάλυση μονοπατιών
- Προβλεπτικά μοντέλα
- Δυναμικά μπαγιεσιανά δίκτυα
- Computational methods
- Modeling cancer progression
- Differentially expressed genes
- Pathway analysis
- Predictive modeling
- Dynamic bayesian networks
- Επιστήμες Υγείας
- Ιατρική και Επιστήμες Υγείας
- Ιατρική Πληροφορική
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Βιοπληροφορική
- Health Sciences
- Medical and Health Sciences
- Medical Informatics
- Computer and Information Sciences
- Natural Sciences
- Bioinformatics
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/48203