New methods for protein structure threading and domain identification

Sammanfattning: The availability of three-dimensional structures of proteins has provided enormous insight into the detailed molecular machinery of the living cell. Since the late 1950s and early 1960s, experimental methods for deriving protein structures have contributed more than 30,000 entries to the Protein Data Bank, representing the collected knowledge of protein and other macromolecular structures. However, sequencing techniques developed since the 1970s have identified numbers of protein sequences that far exceed the current number of experimentally solved structures. Providing alternative methods for deducing structural information of proteins has been one of the goals in the field of structural bioinformatics, in an effort to relieve the pressure put upon experimentalists by using computational methods for protein structure prediction, functional annotation and classification. We have addressed some of these issues and present new methods used primarily for protein structure prediction via homology modelling. Current protein databases classify protein domains into groups based on similar structural features, but this work is manually intense and is not able to keep up with rapid growth in available structural data. Therefore, automated algorithms used to consistently define and systematically identify protein domains have been of increasing importance. Proteins domains are often preferred as templates over intact multi-domain proteins in fold recognition and in homology modelling calculations. A single protein can consist of several domains that may appear in a unique combination compared to proteins in the structure database, therefore sequence-based searches against individual domains more likely to be successful. Therefore we have developed a method of identifying protein domains DOMID. We have approached the structure modelling challenge by deriving a method to perform multiple-template sequence to structure alignment for proteins which is particularly effective for target proteins with low sequence identify to known structures using conserved structural features and sequences similarities.