C.M. Weeks

SHAKE-AND-BAKE SUMMER '98 UPDATE: IMPROVING LIMITED DATA WITH MAGNITUDE ESTIMATION

Hauptman-Woodward Medical Research Institute, Inc. 73 High Street, Buffalo, New York 14203-1196 USA
E-mail: weeks@hwi.buffalo.edu

The Shake-and-Bake algorithm [1] is a state-of-the-art formulation of direct methods which alternates reciprocal-space refinement with filtering in real space to impose constraints. As implemented in the program SnB [2], Shake-and-Bake employs a multi-trial approach in which initial trial structures consist of randomly positioned atoms, phase refinement is based on either tangent-formula or parameter-shift optimization, and density modification in the form of peak picking is used in real space. The program is capable of providing solutions for structures containing as many as 1000 independent non-H atoms (e.g., triclinic lysozyme) and for Se substructures containing as many as 30 anomalously-scattering sites. Readers interested in more information about SnB are referred to Web site www.hwi.buffalo.edu/SnB.

The requirement for relatively high-resolution data, typically 1.1-1.2 A (except for substructure data where 3.0 A is more than adequate), continues to limit the applications of all direct methods, including Shake-and-Bake. The summary of a recent study of three P1 protein data sets (vancomycin, alpha-1 peptide, and lysozyme) is presented below. This study considered the effects of accuracy, completeness, and resolution on Shake-and-Bake success rates (percentage of trial structures refining to solution). Comparison of the results for the experimental data with those for error-free data sets computed using the known atomic coordinates illustrates the importance of collecting complete data to the highest possible resolution.

Success Rates for 500 Trials
Structure	Complete-ness	Resolution	Experimental	Error-Free	Error-Free Complete	Error-Free Extended to 0.85 A

Vancomycin	80.2%	0.97A	< 0.2%	< 0.2%	14%	80%
Alpha-1 Peptide	85.6	0.90	14	19	29	42
Lysozyme	68.3	0.85	< 0.2	< 0.2	0.6	---

The results for vancomycin experimental data could be improved if additional (unmeasured) reflections, likely to have large normalized structure-factor magnitudes, were i) first identified based on probabilistic criteria and ii) then incorporated into the Shake-and-Bake procedure. The success rate for 'complete' experimental data at 0.97A was 2.6%, and this rate was increased to 6.3% when the 'complete' data were extended to 0.85A.