DATA REDUCTION AND SCALING FOR MAD (WITH SOME COMMENTS ON PHASE DETERMINATION). Alan M. Friedman, Dept. of Biological Sciences, Purdue University, West Lafayette, IN 47907

During our work on the crystal structure determination by MAD of the lac repressor core fragment and the phage T4 gene 32 protein: ssDNA complex, we found it necessary to develop a new software package for MAD data reduction, scaling and phase determination. This package, called MADPRB, is derived from the package of Hendrickson, and has been employed in the successful structure determinations above and subsequently in other structures. In many cases MADPRB succeeded where other packages failed. We will discuss several of these cases and describe the principles behind data reduction and scaling in MADPRB.

The scaling program of MADPRB (NEWLSC) starts with a set of unreduced hkl's and integrated intensities from individual observations that have been collected to yield Bijvoet pairs of measurements at several wavelengths that are matched for absorption and decay. Using a moving box local scaling algorithm, the matched measurements are scaled so as to reduce the errors in the Bijvoet and dispersive anomalous differences. In order to do this the user need only input the relationship between matched Bijvoet pairs (e.g. inverse beam). From this the program understands the data collection geometry and automatically scales each observation with its matched mates and sorts all the data into sets for phase determination. The utility of this scaling in reducing errors will be demonstrated.

I will also make some comments about phase determination for MAD. I will suggest that the best method of phase determination requires calculating phases from the matched sets of observations, rather than by "pseudo-MIR" approaches which merge the raw data. This strategy has been adopted in MADPRB. Phase determination in MADPRB occurs in two passes. In the first pass, the program MADRBE estimates the MAD parameters, Fa, Fz and delphi by a modified "algebraic method." Several modifications improve the stability of the estimation for weak and/or poor data. The first pass of phase determination is completed by heavy atom refinement and total phase calculation as in the original Hendrickson package. In a second pass, a new program (BAYESFA) determines phases once the heavy atom parameters are refined. The programs are under further evolution, while the current version is available from the author (email: afried@bilbo.bio. purdue.edu) (Supported in part by NIH GM22778 (to T.A. Steitz) and NSF MCB 9527131 (to A. M. Friedman)