W0229

Blind-Model Refinement of Macromolecular Phases for High Resolution Data. David A. Langs, Hauptman-Woodward Medical Research Institute, 73 High Street, Buffalo, NY 14203 USA

A method has been devised to refine high resolution macromolecular structures without having consciously to fit a model to an electron density map. The procedure begins with a list of the strongest peaks from the initial starting E or F-map. E-values are computed from these peaks using their interpolated electron density peak height as a point-atom atomic form factor. Peaks are then eliminated if they cause a decrease in the correlation coefficient between E-calc and the scaled E-obs for the complete data set. The remaining peaks are used to calculate phases for an ever increasing fraction of the list of F-values which are sorted in decreasing order of magnitude of E-obs. The next F-map is computed and this procedure is recycled five or ten times, each time increasing the number of peaks and the fraction of data, until all data are included and convergence is achieved.

Results for a number of 1.0 Å data sets are very encouraging. Typically, the mean phase error of the initial map may be reduced from about 45 to 15 degrees in five or ten refinement cycles. This usually requires only 5 or 10 minutes of computing on an SGI Indigo workstation. The 15 degree maps have essentially the same features as the final structure-refined maps that required one to initially fit and correct the structural model at each stage of the structure refinement. This often requires a good number of days of skilled labor by a dedicated worker. Much of the trial and error involved in fitting the initial model can thus be avoided by the new procedure.

Research support from NIH grant GM-46733 is gratefully acknowledged.