This map improvement server returns a bias minimized, 6-fold averaged map generated from a model and diffraction data (with optional preceding Molecular Replacement). It does not build or repair the model for you (yet). For automated model building, you need to install a local copy of CCP4 and ARP/wARP (aka wARP&Trace), RESOLVE, MAID, or TEXTAL.

- A reflection data file with data to at least 2.8A. Accepted formats are ScalePack (*.sca), XPLOR/CNS (*.cv) and generic (3I4, 2F8.2) h,k,l,f,sig(f) files (*.fin). The files should not contain anomalous data, and we cannot guarantee the conversion of each and every XPLOR/CNS file. The input validation program will let you know whether it succeeded and probably indicate the reason for failure.
- If any other format than ScalePack is selected, cell constants and allowed spacegroup in CCP4 format must be selected. Inconsistencies between space group and crystal system will trigger a warning and should be verified.
- A structure model in PDB format. This can be a MR search model or a model at any stage of refinement. The model should have proper chain identifiers. They are needed to create proper real space correlation plots. Most errors in pdb files are corrected, but we cannot anticipate every possible liberty taken. Waters can be automatically removed, other parts you do not want to use in MR must be removed.
- A few control parameters, with most defaults set.

- Improvement of map quality at any stage of refinement and model building.
- Phase extension from low resolution model and higher resolution data (superior to difference methods).
- Validation of ligands or solvent by recovering improved electron density (omit ligand in input pdb file).
- Fast search for bound drugs in model after quick MR search.
- Publication quality real space correlation plots to convince readers or reviewers with unbiased local quality assessment of your structure without the need to submit coordinates and data. B-factor plots and B-factor histograms are also returned.
- Convert data or make plots only.

The program does not have any means to create 'de novo' phase information. A balance between starting phase quality and data quality must be maintained: You can either have weak (low res) data, and a reasonable starting model, or good (high res) data, in which case even a poor model can produce a very good map. Example: MR solutions with correlation coefficients as low as 0.32 were successfully processed, given 1.7 A data. Bad data and bad model will not work. The reasons are lack of information (GIGO-principle) ultimately causing limited stability of unrestrained ML refinement of the dummy atom model. If you are unsure about the preformance in severe borderline cases (for example small, densely packed molecules and bad data and bad model), remove known model parts and observe whether they are recovered, as well as observe whether density ignores questionable model parts. Both cases indicate that bias was removed. The real space correlation plot will also reflect this via distinct and clear deviations (negative spikes) in questionable parts of the model. A relatively flat real space correlation combined with low mean correlation (in the 0.6 range or worse) nearly always indicates failure of the method. This model probably cannot be saved this way, perhaps try RESOLVE - although it probably will also yield to the GIGO principle. Generally, the performance of the program is quite convincing, and we have produced numerous improved quality models for many labs.

Users must apply for an account to submit jobs. After submitting a job, initial validation takes place, and the user is prompted whether to continue executing the job. Once the user confirms, an e-mail notice is sent, and it takes one to twenty hours, depending on the complexity of the problem and server workload, for results to become available. In the meantime, you will be able to see (or download, if you are curious) the temporary files and logs while the job is progressing. If the job succeeds, you will receive an e-mail confirmation and be able to retrieve the following

• A *.phs file (h,k,l F, FOM, PHWT) to create a bias minimized map with f*fom and best phase as Fourier coefficients (best viewed in XtalView, do not forget to select f*fom as map type).


• The re-placed input model (sorry no automated model building yet).


• Real space correlation plots beween the fit of the model to the electron density as well as B-factor plots. Follow this link for an explanation of how to interpret real space correlation plots.


• The data file converted into *.sca, *.fin, and mtz format (5% free flags set).


• The phased data in MTZ format (HKL, FP, SIGFP, FOM, PHWT, FreeR_Flag)