Improvement of SAS triple invariant estimates for macromolecular direct-methods phasing

Single-wavelength anomalous dispersion (SAS) data can in principle be phase d by direct methods since a priori estimates of the three-phase structure i nvariants can be computed from these data. The mean phase error of the most reliable triple estimates for a small protein, however, is typically no be tter than 60 degrees, and does not bode well for applications to larger str uctures. A procedure is described that can substantially lower the error in these estimates and introduce a larger number of useful triple invariants into the phasing process. The mean phase error of the most reliable triples for a 2.5 Angstrom resolution data set from a Pt derivative of a 115-resid ue protein was reduced from 55 to 25 degrees by this method. It was also po ssible to identify a significant number of the poorest triple estimates, th ose with mean phase errors approaching 90 degrees, such that they could be reliably down-weighted or excluded from the phasing process.