next up previous contents
Next:  The AUTO wrapper. Up:  Using the program Previous:  Supported crystallographic calculations   Contents

 The .mtz wrapper

GraphEnt offers limited capabilities for a completely automated run with only input the name of a .mtz file. The major limitation is that you do not assign columns or chose a type of calculation. What happens is that GraphEnt will open your .mtz file and read the list of column types. If it finds a recognisable set of column types it will simply go ahead and do the calculation. If what GraphEnt decides to do is not what you wanted, simply use mtzutils or sftools to select, or create the column types that GraphEnt expects for your type of calculation.

This is the list of column types ( order is important) and corresponding calculation that GraphEnt will perform :

Column types Action performed
HHHFQPW Assumed to be h, k, l, F,sigma(F),phi, FOM. Synthesis will be mFexp(iphi)
HHHFQP Assumed to be h, k, l, F,sigma(F),phi. Synthesis will be Fexp(iphi)
HHHFQFQDQ Assumed to be h, k, l, FP,sigma(FP), FPH,sigma(FPH),DFano,sigma(DFano) for a derivative. Synthesis will be a (FP - FPH)2 isomorphous difference Patterson function. An input file for the anomalous synthesis will also be prepared (which can be used as input for a second run).
HHHFQFQ Same as the previous one, but no additional input file for the anomalous part is prepared.
HHHDQ Assumed to be h, k, l,DFano,sigma(DFano). Synthesis will be an anomalous difference Patterson function ( DFano2).
HHHFQ Assumed to be h, k, l, F,sigma(F). Synthesis will be the Patterson function.

Please note that GraphEnt will only check the column types immediately after the indeces, and if a match is found the rest of the columns will be ignored. Furthermore, the search is performed in the order shown in the table and the calculation performed is the first matching. What this means is that if your column types are HHHFQP, GraphEnt will go for a phased synthesis no matter what you may wanted to do. If your intention was to calculate a Patterson function with Fs and sigma(F)s, you will have to use mtzutils or sftools to remove the column containing the phases.

If the column types of your .mtz file are in the right order, just give GraphEnt <my_file.mtz> for a run that will use all data present in the file, or GraphEnt 15.0 3.5 <my_file.mtz> to use only data between 15 and 3.5Å resolution. If what you are calculating is isomorphous difference Patterson functions for a derivative, and your space group has centric zones of the type h0l, hk0, 0kl, you may as well try something like GraphEnt h0l 15.0 3.5 <my_file.mtz> to calculate the [010] Patterson projection. GraphEnt will also recognise and use all zone selections recognised by mtzutils (ie H00, 0K0, 00L, HH0, -HH0, HHH, HK0, 0KL, H0L and HHL).

NOTE WELL : GraphEnt is always performing the calculation in space group P1. To avoid unnecessary repetition, the program calls CAD (and possibly MTZUTILS) from the CCP4 distribution to expand the data to P1. This means that GraphEnt will fail if CCP4 is not correctly set-up or if the various symbols are not defined (especially the CLIBD variable, you can check its presence with setenv | grep 'CLIBD'). NOTE WELL : Maximum entropy maps may well predict non-zero amplitudes for data beyond the high resolution limit of your input data set (thus giving --for good data-- a degree of ``super-resolution''10). For this reason the grid size that GraphEnt uses is significantly larger than that used by the conventional FFT (and even this may not be large enough). The implication is that if you want to do a calculation using all your data to 2Å resolution, you are better-off submitting a batch job for the night instead of trying to do it interactively. In addition, and because GraphEnt is doing the calculation in P1, the higher the symmetry, the larger the grid, the slower GraphEnt will be. As a last precaution, I should add that I have never performed a calculation with more than 9437184 (=192x256x192) pixels.


... ``super-resolution''10
Hopefully, and if the F000 and standard deviations of your data are correctly estimated, the maxent map should only show you the degree of resolution that is required by your data. The fact that it may look sharper than the conventional map is not due to a ``super-resolution'' effect arising from the maxent algorithm, but because the conventional transform is not optimal for your problem. To make this clear, if you had a 100% complete and noise-free data set extending to infinite resolution, then the conventional and maxent map would be identical. Actually, if you start using maxent systematically, you will note that the conventional and maxent maps start looking quite similar under considerably less stringent conditions.

next up previous contents
Next:  The AUTO wrapper. Up:  Using the program Previous:  Supported crystallographic calculations   Contents
NMG, Nov 2002