Results - Details
Search command : Author="Νικολάου"
And Author="Μαρία"
Current Record: 7 of 18
|
|
| Identifier |
000423645 |
| Title |
Identification of gene expression signatures and patterns of disease and treatments in rheumatoid arthritis |
| Alternative Title |
Αναγνώριση μοτίβων γονιδιακής έκφρασης και γονιδιακών υπογραφών μεταξύ ασθένειας και θεραπειών στη ρευματοεϊδή αρθρίτιδα |
|
Author
|
Τσοχατζίδου, Μαρία
|
|
Thesis advisor
|
Νικολάου, Χριστόφορος
|
|
Reviewer
|
Τοπάλης, Π.
Μπερτσιάς, Γ
|
| Abstract |
This work is an attempt of identification of gene signatures and patterns in rheumatoid arthri¬tis and widely used anti-TNF treatments. The experimental procedure included wild-type (healthy), huTNF-transgenic (diseased) and huTNF-transgenic mice treated with one anti-TNF treatments: infliximab, adalimumab, etanercept or certolizumab pegol with a total of 63 sam¬ples. Total RNA was isolated from aqueous extracts of the animals' whole ankle joints which where analyzed with Affy Mouse Gene 2.0 standard array. After data pre-processing two bi-clustering algorithms, Plaid (Turner et al., 2005) and ISA (Csardi et al., 2010), were applied, in order to locate sub-matrices of the initial dataset with distinct expression patterns between different combinations of samples. Plaid is a distribution parameter identification method which was combined with an ensemble method to take into account strict and looser thresholds (Kaiser and Leisch, 2008, Kasim et al., 2016) as well as multiple initialization seeds. ISA is a greedy algorithm which is initiated from an input seed that corresponds to a set of genes or samples. The set is improved at each iteration by adding and/or removing genes and/or samples un¬til convergence is reached with a stable set that is evaluated through correlation of rows and columns. Goal of this procedure was to link treatments and conditions through common expres¬sion patterns of the genes that participated in each group and extract interesting functions from these modules. The results showed that mostly mild patterns were identifiable in our dataset. Some interesting functions also emerged, but sample representation was inadequate in most of these modules either due to the strict analysis applied or due to unsuitability of the method for the current data.
Our next goal, was to identify a unique group of genes among each of the 4 treatment and the two conditions (Tg and Wt) that would maximize the distance between Tg samples' expres¬sion profiles and huTNF-transgenic treated samples while minimizing the distance between the latter and the Wt samples. For this reason, an implementation of a genetic algorithm (GA) (Holland, 1975) approach was designed, which forms subs-groups of possible solutions each one called "chromosome", evaluates them through a fitness function and produces new solutions based on the Darwinian concept of evolution. The advantage of this procedure is the flexibility of the fitness function. In our case, a version of Dunn index was used which is widely applied as a measure of clustering quality. Its advantage is that takes into account distances from both centroids (Wt and Tg in our case). After multiple independent runs, we concluded that the al¬gorithm converges to multiple sub-optimal solutions that are similar only in terms of fitness and not in terms of genes that participate. Thus, this approach was not dependable when applied on the current dataset.
|
| Language |
English |
| Issue date |
2019-07-17 |
| Collection
|
School/Department--School of Medicine--Department of Medicine--Post-graduate theses
|
|
|
Type of Work--Post-graduate theses
|
| Permanent Link |
https://elocus.lib.uoc.gr//dlib/f/4/f/metadata-dlib-1572863802-878983-15146.tkl
|
| Views |
192 |
Search
