Correlated Phasing of Multiple Isomorphous Replacement Data

Abstract

Substantial highly correlated differences sometimes exist between a series of heavy-atom derivatives of a macromolecule and the native structure. Use of such a series of derivatives for phase determination by multiple isomorphous replacement (MIR) has been difficult because MIR analysis has treated errors as independent. A simple Bayesian approach has been used to derive probability distributions for the phase in the case where a group of MIR derivatives have correlated errors. The utility of the resulting 'correlated-phasing' method has been examined by applying it to both simulated and real MIR data sets that contain sizeable correlated errors and it has been found that it can dramatically improve MIR phase estimates in these cases. Correlated phasing is applicable to situations where derivatives exhibit substantial correlated changes in protein conformation or crystal packing or where correlated errors in heavy-atom models are large. Correlated phasing does not substantially increase the complexity of phase computation and is suitable for routine use.