Free R value: cross-validation in crystallography.

Publisher Summary This chapter focuses on the free R value and other applications of cross-validation in crystallography. Powerful methods have been developed to lower the chances of misinterpreting or overinterpreting diffraction data. Some of the more important methods measure the agreement of the structure with empirical rules about protein folds, comprehensive conformational analyses, the real-space correlation coefficient, and the free R value. The protein-folding rules and conformational analyses depend on empirical knowledge of protein structure. They validate the model regardless of the fit to the diffraction data. In contrast, the real-space correlation coefficient and the free R value are entirely diffraction data based and are applicable to any macromolecule. They validate the extent to which the model explains the diffraction data. Even the best bulk solvent and multiconformer models result in free R values of around or slightly above 20% for the penicillopepsin crystal structure. A free R value of 20% is significantly higher than expected from the estimated high statistical quality of the diffraction data. Thus, it is conceivable that the present models for solvation and thermal motion are incomplete. Alternatively, the intensity data might be affected by systematic errors of unknown origin. These facts point to the need for the solution of benchmark macromolecular structures at high resolution and with accurate experimental phases.