Intensity-based domain refinement of oriented but unpositioned molecular replacement models

A program is described that performs least-squares group refinement of oriented molecular replacement models whose positions in the unit cell are unknown. The program (INTREF) is designed to produce improved models for use in a translation function by optimizing the orientations and relative translations of the model domains. The molecular contents of the asymmetric unit are refined as a small number of rigid bodies whose origins relative to each other may be unknown. More than one molecule in the asymmetric unit can be accommodated. The refinement seeks to minimize the residual error between the observed and calculated intensities that have been modified to produce the equivalent of a radial weighting in Patterson space. Calculated intensities include contributions from all symmetry-related molecules, enabling meaningful refinement in high-symmetry space groups. Derivatives of the intensities with respect to the rigid-body parameters are evaluated numerically using fast Fourier transforms and the shifts are obtained by non-linear least-squares analysis. Results with test cases show that the program is capable of adjusting the orientations and relative translations of protein domains to give models that more closely resemble the known structures. Consequently, the resulting models produce more accurate and more interpretable results in translation functions. The importance of including all crystallographically related molecules and of downweighting the contribution of the longer-radius region of the Patterson function is demonstrated.