In Chapter two methods of calculating structure factor phases from MAD data
were described. The classical phasing approach incorporated into the program MLPHARE was used
here for phase determination. The algebraic approach however has not been attempted for the reasons
that follow. The MADSYS suite of programs which implement the algebraic approach have been
specifically developed for use with MAD data measured in one of two ways; i) with Bijvoet mates
recorded almost simultaneously on the same or adjacent images, or ii) using inverse beam geometry.
The advantage of inverse beam geometry is that the
recorded Friedel mates will have suffered from similar absorption effects [54] thus
minimising the contribution from that source of systematic error. The data measured during this
work was not collected in this fashion and was therefore not suitable for treatment with the MADSYS suite as it stood. A comparison of the two methods of phase determination has been given by
Ramakrishnan *et al* [95]. The structure of the globular domain of histone H5
was determined by MAD to 2.5Å resolution using the selenomethionyl protein. Two Se atoms were
present per 89 amino acid residues. The correlation coefficients against the final refined structure
were calculated for the two resulting electron density maps and suggested that the MLPHARE
phased map was of a higher quality with a correlation coefficient of 0.54 compared with 0.43 for the
MADSYS phased map. After solvent flattening the correlation coefficients of both maps
increased to 0.63 (MLPHARE) and 0.56 (MADSYS). One advantage of the MLPHARE
procedure in the histone H5 case appeared to be that phases could be determined for of the
recorded data compared with only phased by MADSYS.

One characteristic of the MLPHARE program which is apparent from the results presented in the previous chapter is the unmeaningful figure of merit values. The magnitudes of the FOM's appeared to give a false indication of the overall quality of the phasing solution. This is most apparent in the case of the SIRAS phasing solution where the highest observed FOM's result in a very poor correlation coefficient. The failure of the FOM value to act as a consistent test of a MAD solution may lie with the calculation of the combined phase probability distribution for a number of derivatives (Eq. in Sec. ). This treatment of probability distributions is valid only when the individual distributions are statistically independent. This assumption may be valid for MIR phasing, where the distributions are representative of experimental errors in structure factors measured from derivatives where the heavy atoms differ in size and occupy different positions in the unit cell, but is evidently weaker for MAD phasing where the heavy atom sites are common in all measurements. This may be demonstrated simply by running the phasing program and including the same derivative two or more times. Although no new information is being added into the procedure, it is often observed that the FOM increases. This is unfortunate as it leaves the user with no universal quantitative guide as to the quality of the solution. The various values of the Cullis R-factor on the other hand appeared to provide a consistent idea of the quality of individual data sets although their values may still not give any information about the quality of the overall result. The assessment of the solutions presented here has therefore depended solely on average phases differences and correlation coefficients between the experimental and the `target' electron density maps, an option which is not open when a new structure is being researched. It would be useful for the future to establish a more realistic figure of merit which could be applied to MAD phased data. The problem of solving a protein structure ultimately involves the interpretation of an electron density map and relies therefore very much on subjective judgement, especially when the maps are of low quality with poor connectivity. The improvement of phasing methods and the use of some figure of merit for phasing solutions is aimed at making the job of map interpretation as straightforward as possible.

The refined values of the anomalous scattering factors showed remarkable agreement with those predicted from fluorescence measurements with r.m.s. deviations about the experimental curve of 1.36e for and 0.87e for . This result suggests that precise knowledge of the absolute values of the anomalous scattering factors is not a pre-requisite for MAD phasing, however the determination of the form of the anomalous scattering factors about an absorption edge is necessary as it permits the correct choice of X-ray energies for the MAD experiment to be made.

Fri Oct 7 15:42:16 MET 1994