Quantitative study of the effects of chemical shift tolerances and rates of SA cooling on structure calculation from automatically assigned NOE data

The calculation of protein structures from nuclear magnetic resonance (NMR) data has been greatly facilitated by improvements in software for the automatic assignment of NOESY spectra. Nevertheless, for larger proteins, resonance overlap may lead to an overwhelming number of assignment options per peak. Although most software for automatic NOESY assignment can deal with a certain level of assignment ambiguity, structure calculations fail when this becomes too high. Reducing the number of assignment options per peak by reducing the chemical shift tolerances can lead to correct assignments being excluded, and thus also to incorrect structures. We have investigated, systematically, for three proteins of different size, the influence of the chemical shift tolerance limits (Δ) and of the number of simulated annealing (SA) cooling steps on the performance of the software ARIA. Large tolerance windows, and the correspondingly high levels of ambiguity, did not cause problems when appropriately slower cooling was used in our SA protocol. In cases where a high percentage of well-converged structures was not achieved, we demonstrate that it is more productive to calculate fewer structures whilst applying slow cooling, than to calculate many structures with fast cooling. In this way, high-quality structures were obtained even for proteins whose NMR spectra showed great degeneracy, and where there was much inconsistency in peak alignment between different samples. The method described herein opens the way to the automated structure determination of larger proteins from NMR data.