Aristotle University Of Thessaloniki (AUTH)
Ανάπτυξη ενοποιημένων τεχνικών διαχείρισης και επεξεργασίας βιολογικών δεδομένων με εφαρμογές στη μελέτη της γονιδιακής έκφρασης
Abstract
dc:descriptionThe main objective of this thesis is the development of integrated methods for managing and processing biological data with specific emphasis on their application to computational prediction of gene structures in eukaryotes. Most biological analysis tools developed so far are characterized by multi-level heterogeneities that make combinatorial usage and further analysis difficult and problematic. Scientific workflow management systems offer integrated environments for tools orchestration in successive steps, though the selection of the best-fitting tool in each step remains an important issue, considering that the outcome of the underlying computational models are very frequently differentiated. The integration architecture proposed in this thesis offers transparent access to publicly available tools that fulfil common functions, enabling comparative post-analysis of their outcomes. Specifically, the proposed architecture consists of: a) Appropriate wrapping/parsing modules, b) a common schema for describing the results of the predictive modeling, c) combinatorial visualization modules, and d) query formation and execution modules that apply on multiple commonly-described outcomes. The applicability of the architecture was evaluated on a set of ab initio gene predictors. The modular design of the architecture allows for additional functionalities to be implemented, as well as for the incorporation of supplementary schemas describing relevant tools. In this context, the architecture was extended by embodying schema descriptions of publicly available tools that predict specific signal sensors. Signal sensors define the boundaries of functional tracts within a genomic region and their computational prediction, coupled with the outcome of gene structure predictors can be used to increase the efficiency of the underlying learning methods. Splice sites are important signal sensors that define the synthesis of the protein product. Splice sites are located at the beginning and end of an intron and signal the coding regions that are going to be translated into proteins. The biological mechanism that recognizes splice sites involves multiple, complex interactions among adjacent and non-adjacent nucleotides. Our deficient knowledge of these interactions put obstacles in predictive modeling of splice sites. This thesis presents a hybrid method for predicting splice sites that consists of two successive classification steps. The first step is undertaken by a Gaussian support vector machine that is trained on probabilistic data descriptions, using different feature selection methods. The second step combines the evidence of specific features resulted from relevant published studies with the probability estimates of the first classification step, in order to induce a binary decision tree. Finally, the thesis proposes different analysis types of alternatively spliced exons and their neighboring intronic regions, in order to investigate potential discriminative features that are differentiated between constitutive and alternative gene expressions. The results of the analysis give important evidence that is biologically useful, while from the computational point of view they could be used to feed the proposed hybrid identification method, in order to predict alternative splice sites.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2008
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Malousi, Andigoni
- Μαλούση, Αντιγόνη
Subjects
dc:subject × 20- Βιοπληροφορική
- Γονιδιακή έκφραση
- Ενοποίηση εργαλείων
- Μηχανική μάθηση
- Εναλλακτικό μάτισμα
- Πρόβλεψη γονιδίων
- Bioinformatics
- Gene expression
- Tool integration
- Machine learning
- Alternative splicing
- Gene prediction
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Ιατρική και Επιστήμες Υγείας
- Επιστήμες Υγείας
- Natural Sciences
- Computer and Information Sciences
- Medical and Health Sciences
- Health Sciences
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/18716