AN IMPROVED CROSS-VALIDATION METHOD DENSITY MODIFICATION. APPLICATION TO PHASE REFINEMENT AND EXTENSION. J.-S. Jiang, A. T. Brünger, The Howard Hughes Medical Institute and Department of Molecular Biophysics and Biochemistry, Yale University, New Haven, CT 06520, USA

The cross-validated or free R value is correlated with the phase accuracy of an atomic coordinate model in the course of model building and refinement (Brünger, 1993, Nature, 355, 472-474). Application of cross-validation to phase improvement by density modification causes difficulties in the absence of an atomic model (Roberts & Brünger, 1995, Acta Cryst., D51, 990-1002). The underlying reason can be found in the removal of partial information from the iterative electron density map calculation required in density modification schemes. This problem can be partially overcome by complete cross-validation albeit at a significant computational expense.

An improved cross-validation scheme with a single test set is proposed for phase refinement and extension in density modification schemes, such as solvent flattening, histogram matching, Sayre's equation, skeletonization, and non-crystallography symmetry averaging. It is based on "interpolating" the test set when computing electron density maps from working sets. This interpolation method which is commonly used for density modification when only partial diffraction data are available appears to improve the reliability of the free R value for density modification. The free R value computed with the new scheme shows more correlation to mean phase error than the original scheme when a single test set is used.