Graph-Based Approaches to Protein StructureComparison - From Local to Global Similarity
The comparative analysis of protein structure data is a central aspect of structural bioinformatics. Drawing upon structural information allows the inference of function for unknown proteins even in cases where no apparent homology can be found on the sequence level. Regarding the function of an en...
|PDF Full Text
No Tags, Be the first to tag this record!
|The comparative analysis of protein structure data is a central aspect of structural bioinformatics. Drawing upon structural information allows the inference of function for unknown proteins even in cases where no apparent homology can be found on the sequence level.
Regarding the function of an enzyme, the overall fold topology might less important than the specific structural conformation of the catalytic site or the surface region of a protein, where the interaction with other molecules, such as binding partners, substrates and ligands occurs. Thus, a comparison of these regions is especially interesting for functional inference, since structural constraints imposed by the demands of the catalyzed biochemical function make them more likely to exhibit structural similarity. Moreover, the comparative analysis of protein binding sites is of special interest in pharmaceutical chemistry, in order to predict cross-reactivities and gain a deeper understanding of the catalysis mechanism.
From an algorithmic point of view, the comparison of structured data, or, more generally, complex objects, can be attempted based on different methodological principles. Global methods aim at comparing structures as a whole, while local methods transfer the problem to multiple comparisons of local substructures. In the context of protein structure analysis, it is not a priori clear, which strategy is more suitable.
In this thesis, several conceptually different algorithmic approaches have been developed, based on local, global and semi-global strategies, for the task of comparing protein structure data, more specifically protein binding pockets. The use of graphs for the modeling of protein structure data has a long standing tradition in structural bioinformatics. Recently, graphs have been used to model the geometric constraints of protein binding sites. The algorithms developed in this thesis are based on this modeling concept, hence, from a computer scientist's point of view, they can also be regarded as global, local and semi-global approaches to graph comparison. The developed algorithms were mainly designed on the premise to allow for a more approximate comparison of protein binding sites, in order to account for the molecular flexibility of the protein structures. A main motivation was to allow for the detection of more remote similarities, which are not apparent by using more rigid methods. Subsequently, the developed approaches were applied to different problems typically encountered in the field of structural bioinformatics in order to assess and compare their performance and suitability for different problems.
Each of the approaches developed during this work was capable of improving upon the performance of existing methods in the field. Another major aspect in the experiments was the question, which methodological concept, local, global or a combination of both, offers the most benefits for the specific task of protein binding site comparison, a question that is addressed throughout this thesis.