Statistical evaluation of simulated NMR data of flexible molecules

Abstract

A new probability score—named χ-probability—is introduced for evaluating the fit of mixed NMR datasets to calculate molecular model ensembles, in order to answer challenging structural questions such as the determination of stereochemical configurations. Similar to the DP4 parameter, the χ-probability is based on Bayes theorem and expresses the probability that an experimental NMR dataset fits to a given individual within a finite set of candidate structures or configurations. Here, the χ-probability is applied to single out the correct configuration in four example cases, with increasing complexity and conformational mobility. The NMR data (which include RDCs, NOE distances and ³J couplings) are calculated from MDOC (Molecular Dynamics with Orientational Constraints) trajectories and are investigated against experimentally measured data. It is demonstrated that this approach singles out the correct stereochemical configuration with probabilities more than 98%, even for highly mobile molecules. In more demanding cases, a decisive χ-probability test requires that the datasets include high-quality NOE distances in addition to RDC values.