Inferring protein ensembles directly from NOESY spectra

Solution NMR spectroscopy enables precise measurements of proteins in their native-like state. These measurements reveal conformational diversity as ensemble averages. However, conventional NMR structure determination often converts experimental data into restraints for molecular dynamics, which encode mean structure information but fail to retain conformational distribution details.

Ensemble selection, an alternative method, compares experimental observables directly with candidate conformers, allowing for population distribution inference. Despite the potential of NOESY data, the richest structural information source in protein NMR, to infer ensembles, most methods have not incorporated this data due to difficulties in comparing experimental and back-calculated spectra.

The CoMAND method was previously developed to tackle this challenge, demonstrating the practicality of quantitative agreement for NOESY spectra using tailored heteronuclear editing schemes. Building upon this foundation, the researchers have now expanded the approach into a framework capable of directly inferring protein ensembles within a versatile ensemble-selection architecture that integrates multiple NMR observable classes.

A quantitative scoring framework for comparing experimental and back-calculated observables was introduced, combined with regularized ensemble selection and Monte Carlo simulated annealing. This framework was successfully integrated with the OpenMM molecular dynamics engine, enabling the generation of conformational pools using conventional molecular simulation methods.

The approach was applied to human ubiquitin, yielding an ensemble that simultaneously agrees with NOESY, residual dipolar coupling, and scalar coupling data while preserving conformational diversity supported by experimental observations.