AbstractThe research work presented in this thesis stands at the crossroads of computer sciences and biology, as it consists in the development of a computer program to extract useful information from biological data, using an algorithmic strategy inspired by the biological model of the evolution of the species.
Biclustering of gene expression data means analyzing large amounts of biological experimental data mea- suring the level of expression of some genes under some experimental conditions (different tissues, different patients,...), in order to individuate groups of genes that exhibit coherent behaviors under some groups of conditions. Such correlations may provide a hint over an existing biological relation between the indi- viduated genes and conditions, and thus be of particular interest from a biological and medical point of view.
Evolutionary computation is a generic class of algorithms, sharing the use of mechanisms inspired by darwinian evolution to solve problems whose best solutions, according to some fixed quality criteria, have to be discovered. Evolutionary computation techniques can be particularly relevant to achieve an efficient biclustering of gene expression data, and several evolutionary biclustering approaches have been proposed in the literature.
In this thesis, we present MOBPEOC, a new evolutionary biclustering approach that we developed to improve the biclustering mechanism of an existing evolutionary approach. In particular, MOBPEOC represents a first life-size test for a new general-purpose evolutionary technique that we propose for the first time in this work, called probabilistic encoding.
An experimental evaluation of the MOBPEOC algorithm is proposed, where the technique is applied to real biological data. The comparison of the obtained results with the previous biclustering evolutionary approach shows a strong improvement of the quality of the discovered solutions.
|Date of Award||29 Sep 2010|
|Supervisor||Wim Vanhoof (Supervisor)|
- data mining
- gene expression data
- evolutionary compu- tation
- multimodal and multi-objective optimization
- probabilistic encoding
A multi-objective genetic algorithm for biclustering of gene expression data with probabilistic encoding and overlapping control
Marcozzi, M. (Author). 29 Sep 2010
Student thesis: Master types › Master in Computer science