research papers
Macromolecular crystallography increasingly pushes the limits of the size and complexity of the molecules and assemblies under study. Conformational variability in these assemblies frequently results in limited diffraction or in poorer phase information than that typically obtained from smaller macromolecules. Current methods for solving and refining crystal structures often perform poorly under these adverse conditions and thus new methods must be developed. Atomic-model refinement is particularly sensitive to the low information content of limited diffraction data. Here a multi-start procedure is presented for simulated annealing refinement. After multi-start refinement, poorly fitted regions of the model often display increased variability compared with correctly fitted regions. Structure-factor averaging over the resulting models improves the quality of phases derived from atomic models and reduces model bias. Averaging can be performed even at a minimum Bragg spacing of 3.5 Å, taking into account variability of atomic positions due to errors or intrinsic flexibility even when individual B factors cannot be refined. The average structure factor is thus closer to the true structure factor and should provide a better starting point for estimations of σA values for electron-density-map calculations. Test cases show increased phase quality of the average structure factor. The method is most useful when the initial model is poor or when only moderate-resolution diffraction data are available and may allow meaningful phase improvement through atomic-model refinement where it was not previously possible.