Anecdotes...Anecdotes...Comparison of Denzo and Mosflm
I did a side-by-side comparison of procesing exactly the same frame data with denzo/scalepack and mosflm/scala for a 5-wavelength MAD data set. About 1000 frames were reduced with either denzo or mosflm, taking care with each program to carefully refine and re-refine every parameter to convergence, and systematically optimizing all user-configurable parameters. For example, in denzo, the profile fitting radius and each measurment box parameter were each systematically varied, and the value giving the highest I/sigI in the final, merged data set was used. Mosflm, however, optimizes its own measurment box, so no manual tweaking was done. The error scale factor and estimated error keywords for scalepack were optimized systematically and automatically by autoscalepack, and the same was done for the mosflm data in scala, using autoscala. It was decided it would be unfair to scale denzo data in scala, because, unlike scalepack, scala does not postrefine the reflection data. Mosflm, however, "pre"-postrefines while it measures spots.
Once merged data sets were obtained, they were used to calculate phases.
The same heavy-atom positions were input into identical mlphare
and dm procedures
on each data set. After the experimental electron density maps were obtained
(below), the two data sets were subjected to automatic model-building in
wARP 5.0. The refinement
statistics from the two automatically-built models are provided below.
| Denzo | Mosflm | |
| 3.9-5.4% | Rmerge | 5.0-5.5% |
| 0.55 | mean FOM | 0.57 |
| 32.2% | DM Rfree | 34.1% |
| 0.2 - 1.5 | P Power | 0.2 - 1.6 |
| 22.1% | Rcryst | 21.9% |
| 31.2% | Rfree | 31.1% |
One can see that there were some differences between the two data-reduction
programs, particularly in the Rmerge values reported by scalepack.
The final, merged data sets were about 3% different, with high-resolution
and high-intensity hkls showing the most discrepancy between processing
programs. This was most likely due to the differences in profile fitting
and the handling of scanner nonlinearities between the two programs. However,
it is clear from the final statistics and maps that the impact of processing
program choice on the final results in marginal.
The difference in ease-of-use between these two programs depends, of course, on wether or not you are familiar with denzo. However, even if you are familiar with denzo, The command "Wedger Elves hurry up and process /data/frames in P212121 >&! Wedger.log &" will, most likely, be done before you are even ready to run scalepack. :)
Back to the Elves Homepage.