Statistical filtering for NMR based structure generation.

The constitutional assignment of natural products by NMR spectroscopy is usually based on 2D NMR experiments like COSY, HSQC, and HMBC. The difficulty of a structure elucidation problem depends more on the type of the investigated molecule than on its size. Saturated compounds can usually be assigned unambiguously by hand using only COSY and 13C-HMBC data, whereas condensed heterocycles are problematic due to their lack of protons that could show interatomic connectivities. Different computer programs were developed to aid in the structural assignment process, one of them COCON. In the case of unsaturated and substituted molecules structure generators frequently will generate a very large number of possible solutions. This article presents a "statistical filter" for the reduction of the number of results. The filter works by generating 3D conformations using smi23d, a simple MD approach. All molecules for which the generation of constitutional restraints failed were eliminated from the result set. Some structural elements removed by the statistical filter were analyzed and checked against Beilstein. The automatic removal of molecules for which no MD parameter set could be created was included into WEBCOCON. The effect of this filter varies in dependence of the NMR data set used, but in no case the correct constitution was removed from the resulting set.