Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/26C9F4B6-A941-4687-B55A-92499824DE08 | ||
| Year | 2015 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Dimitrios-Apostolos Chalepakis-Ntellis, "Expression networks for cancer gene markers", Diploma Work, Σχολή Ηλεκτρονικών Μηχανικών και Μηχανικών Υπολογιστών, Πολυτεχνείο Κρήτης, Chania, Greece, 2015 https://doi.org/10.26233/heallink.tuc.27006 | ||
| Appears in Collections |
Cells contain thousands of organic molecules, such as genes, RNA, proteins andmetabolites which interact in complex ways. The networks provide a powerfulframework to represent these complex relationships and interactions, which areresponsible for various cellular functions with the effects of individual nodes of the molecules. In this thesis is curried out the learning of Bayesian network structure from gene expression data from breast cancer pathology samples. The Bayesian networks provide a neat and compact representation for expressing joint probability distribution and for inference. The representation and the use of probability theory makes Bayesian networks suitable for combining domain knowledge and data, expressing causal relationships, avoiding overfitting a model to training data, and learning from incomplete datasets. Specifically, to learn such a structure, have been used interactions of 77 genes, which is a gene signature associated with the pathology of breast cancer. Structures were constructed separately for both cancer and control samples, while learning structures were made according to the structure learning algorithm K2, considering the discrete and continuous variables. The resulting structures studied in properties of Small-World and Scale-Free, shown in most real world networks. Furthermore, important nodes as well as complexes (according to the algorithm MCODE) and modules (according to the algorithm jActive Modules) were searched out in the structures which were statistically and biologically evaluated. Statistical analysis showed that the networks have the property of Scale-Free, which is consistent with their biological dimension and that there are important hubs, clusters and modules in the networks. The Bayesian networks analysis dynamically highlighted subnetworks with central hubs which offer new knowledge of the biological pathways involved in cancer breast.
Copyright © DIAS 2013
