An assessment of the precision and accuracy of protein structures determined by NMR. Dependence on distance errors.

We tested the dependence of the accuracy and precision of calculated NMR structures on the errors of the distance constraints using sequential simulated annealing and found that: (1) the accuracy of the family of structures depends mainly on the quality of the data, but is no better than about 1 A even if the errors in distance constraints are smaller than +/- 1 A. (2) The precision of the calculated structures, on the other hand, is nearly insensitive to the quality of the data. With present methods, the accuracy of NMR structures is at best of the order of 1 to 2 A, although a precision of 0.4 to 0.7 A is readily attainable. Comparisons with recent studies of this problem also brought out the importance of distinguishing between correct and incorrect definitions of accuracy when reporting numerical estimates. Using an incorrect definition of the term accuracy can lead to an artificially favorable estimate of its numerical value.