Методи за детекција и анализа на сврзните делови од протеинските структури и нивна примена за одредување на функции на протеини

Abstract

The knowledge about the functions of the protein molecules is very important in order to understand and regulate the processes in living organisms. Therefore, the aim of this PhD thesis is to develop new methods for determining the functions of the protein structures. In this thesis, four methods for determining the structural similarity between protein molecules are presented. Also, the performances of these methods are analyzed in details, and additionally these methods are compared with several existing methods for aligning protein structures. By using these methods, the structural similarity between the comparing protein structures could be determined, and they could be used for classifying and annotating protein structures. In this theses, a novel method for protein function prediction is proposed, where the decisions about the annotations of the inspected structure are based on the annotations of its nearest neighbors. Besides annotation based on structural alignment, the protein function prediction could be made by analysis of the characteristics of the binding sites, which are the regions where the inspected structure interacts with another structure. In this thesis several methods for protein binding sites detection are proposed. Besides the methods based on the classical theory of sets, also the methods based on the fuzzy set theory are taken into consideration. In order to improve the prediction power of the models, ensembles are induced, and also by using feature selection and transformation techniques the most relevant characteristics of the amino acids residues are determined in order to find out which features should be considered in the model induction. The proposed methods for protein binding sites prediction are compared with several existing methods used for this purpose. In this thesis, a novel method for protein function prediction is proposed that is based on the local characteristics of the binding sites, as well as the global characteristics of the protein structure. The model induction is made by using several methods for multi-label classification. Finally, a detailed comparison of the two proposed methods for determining protein functions is performed.