Microsoft Word - predicting NMR parameters

The spectral parameters of the nuclear magnetic resonance (NMR) spectra are dependent on the chemical environment around the nuclei, making NMR spectroscopy a powerful method for studying molecular structure and dynamics at the atomic level. Conversely, the spectral parameters can be calculated if one knows the molecular structure. The spectral parameter prediction plays a key role in many applications of computational NMR. This thesis presents two NMR parameter prediction approaches for two different purposes. Chemical shifts are the key parameters of the NMR spectrum. In the field of protein NMR, the use of chemical shifts in protein structural studies has been increasing in the last years, driven by improvements in the chemical shift prediction methods. In addition to the protein structure, chemical shifts are dependent on protein dynamics. In order to account for the dynamic effects, a four-dimensional approach for protein chemical shift prediction was developed. Here, the 4th dimension is time and it is mapped by molecular dynamics simulations. The conformational space was further expanded by starting the MD simulations from different conformations of the same protein. From the structural parameters averaged over the conformational space of the MD simulations, chemical shifts prediction models for all backbone and most side chain nuclei were built with principal component regression. In comparison with the non-dynamic models, the dynamic models achieved 13 % lower root-mean-square (RMS) errors for different backbone nuclei, underlining the importance of dynamics in reproducing experimental protein chemical shifts. An additional outcome of the project is the prediction program 4DSPOT, which is freely available for the protein NMR community (www.uef.fi/4dspot). NMR spectra can be simulated and iteratively analyzed with quantum mechanical principles if the parameters are known. The scalar coupling constants are the parameters that give rise to the fine structure of the NMR signals. The second prediction method presented in this thesis targets small molecule couplings to be used in automatic spectrum analysis. Coverage and speed are emphasized in the design of the method. Thus, the method is based on a lightweight hash dictionary search, followed by a k Nearest Neighbors regression to resolve the substituent and conformational dependencies. Despite the growth of databases, there are still many situations when experimental data is too sparse to permit prediction model building. However, recently the accuracy of quantum-mechanical calculation of NMR parameters has greatly improved. Therefore, the use of quantum chemistry as a source of teaching data is discussed and some preliminary results are shown. National Library of Medicine Classification: QU 25 Medical Subject Headings: Nuclear Magnetic Resonance, Biomolecular; Spectrum Analysis; Molecular Conformation; Proteins; Molecular Dynamics Simulation Library of Congress Subject Headings: Nuclear magnetic resonance spectroscopy; Proteins Structure