Abstract
Protein structures play a pivotal role in understanding protein functions. X-ray crystallography and nuclear magnetic resonance spectroscopy (NMR) are two main experimental approaches for protein structure determination. The structures determined by NMR are not as accurate as those by X-ray crystallography in general, due to limited distance data (e.g. NOEs) that can be generated from NMR experiments. The NMR modeling algorithms are not well developed either, to produce high-quality structures with desired accuracy and efficiency. To address the NMR modeling issues, a so-called geometric build-up approach for NMR modeling has been developed, which remarkably reduces computing time in determining NMR structures. On the other hand, a novel bioinformatics approach aimed at increasing the accuracy of the NMR-determined structures has been proposed and tested, by deriving more distance constraints, in addition to the experimental ones, from the databases of known protein structures. Results show that with database-derived distance constraints, NMR-determined structures can be improved significantly based on various standard evaluations such as acceptance rate, RMSD to corresponding X-ray structures, Ramachandran plot, etc. The derived constraints are also able to enhance and even replace some experimental restraints such as short-range NOEs and dihedral angles, holding a promise of reducing experimental and labor costs. In addition, they can be applied to improving the structure of the under-determined regions of proteins such as the loop regions of the prion protein, a protein responsible to a group of important neurodegenerative diseases including the Mad Cow Disease.