SOLVE scoring
SOLVE uses four scoring criteria to evaluate each solution:
SOLVE uses the 10-30 seeds generated from the analyses of Patterson functions as trial solutions and scores each seed. From this set of scores, an average score and the standard deviation of this average score is obtained for each criteria. A Z-score (number of standard deviations above the mean) is then calculated for each trial solution for each criteria. The overall Z-score for a solution is the sum of the individual Z-scores, corrected for any large deviation among the scores.
Analysis of difference Pattersons
The analysis of the difference Pattersons for a MAD dataset are carried out in two steps. First all Patterson vectors that should result from all the sites in a derivative are calculated. Then the difference Patterson is examined at these positions and the relative occupancies of all the sites are refined so as to match the Patterson as closely as possible. The comparison of observed and calculated peak heights*1000/rms of the map is printed out along with an overall quality of the solution and evaluation of the fit to the difference Patterson.
"Free" self-difference Fourier analyses
The difference fouriers calculated here use all anomalously scattering atoms except just one atom to be examined (and all that are equivalent to it) to calculate native phases. These native phases are used to calculate a difference Fourier; the Fourier is examined at the coordinates of the test site and the peak height/rms of the map is noted. This is repeated for each anomalously-scattering atom in the structure and the heights are printed out.
Non-randomness test on native fourier
Most protein crystals have solvent regions (with low density and low variation in electron density) and protein regions (with high variation of density and moderate average density). This program examines the standard deviation (SD) of the local electron density in various locations around the unit cell and determines if the variation of this SD is high or low. If the phases are random, then you get a low variation of the SD (the whole unit cell is rather uniform and looks like noise). If they are good, you get a high variation of the SD (part of the unit cell is smoother than another part). The program also analyzes the correlation of local r.m.s. density to see whether adjacent regions in the map tend to have similar r.m.s. variation. This is essentially a measure of how contiguous the regions of high and low variation are. This measure is reported back as the "correlation coefficient" (CC) in SOLVE. In versions 1.11 and higher the correlation coefficient is used in the actual scoring procedure (and not the standard deviation).
Figure of merit of phasing
The figure of merit of phasing the native structure proves to be a useful criteria in scoring a solution. Solutions with low figures of merit are unlikely to be correct, while those with a high figure of merit are likely to be right if the other criteria are also favorable.
Correction for uneven scores
SOLVE makes a correction on the score to reflect the discrepancy between the various scores. The effect is to score a solution that has a uniformly good score for all criteria higher than a solution with a very good score on just one criteria.