Refinement with Refmac5
From Media Wiki
Main Page - Using the CCP4 software - Refinement with CCP4 - Refinement with Refmac5
The REFMAC program can carry out rigid body, TLS, restrained or unrestrained refinement against Xray data, or idealisation of a macromolecular structure. The Xray data is normally in the form of observed structure factor amplitudes, although the latest version of Refmac can refine against intensities. If good quality experimental phases are available then they can be used in addition. Refmac minimises the coordinate parameters to satisfy a Maximum Likelihood residual. REFMAC also produces an MTZ output file containing weighted coefficients for SigmaA weighted mFo-DFcalc and 2mFo-DFcalc maps, where "missing data" have been restored.
Contents |
[edit] Running Refmac5
Refmac5 will refine an atomic model by adjusting the model parameters (coordinates, B-factors, TLS etc) in order to obtain the model which best explains the experimental data (i.e. maximises the likelihood). Progress is measured by R-factor and Free R-factor, as well as by the likelihood scores themselves.
To run Refmac, you need at minimum two pieces of information: An MTZ file containing a set of observed structure factors (and optionally phase information), and an initial atomic model from Molecular replacement or Model building.
Launch the 'Refmac5' task in the Refinement module. The task interface (shown on the right) will appear. Now select what sort of refinement you wish to perform; this will usually be 'Restrained Refinement', but you will need to select what type of data to use:
- No phase information: Use this after molecular replacement or at very high resolution.
- Phase and FOM: Use this after experimental phasing only if Hendrickson-Lattman coefficients are not available.
- Hendrickson Lattman coefficients: Use this after experimental phasing.
Enter the names (or browse) for the two files as follows:
- The MTZ file containing your observed structure factors. Enter this in the field labelled 'MTZ in'. You will need to select the MTZ columns for the observed structure factor magnitude (F) and its standard deviation (sigma). If you elected to use phase information, then you should also select the best phase and figure of merit or Hendrickson Lattman coefficients output from the phasing program.
- The initial atomic model from molecular replacement or model building. Enter this in the field labelled 'PDB in'. You should check that Refmac5 understands the chemistry of the contents of the PDB file by running Review Restraints.
Select 'Run now', and the program will run.
[edit] Program output
Double-click your Refmac job in the CCP4 task list to view the log file. This contains both crystallographic statistics and warning messages. Use the 'Find' button to search for the word 'warning' to identify any problems with the input model.
To judge the quality of the refinement model, click 'Show log graphs' and scroll down to the last graph in the list, titled 'Rfactor analysis, stats vs cycle' (see the example on the right). This shows the variation in R-factor and Free R-factor against cycle number. The R-factor and Free R-factor should decrease as the calculation proceeds.
Check the difference between R-factor and Free R-factor. If the Free-R factor is much higher than the R-factor, then you do not have enough data to support the level of detail you are trying to refine. You will need to reconsider the model you are using, for example try using isotropic or overall B-factors instead of anisotropic.
At the end of the log file is a useful table listing R / Rfree, -LL / -LLfree (LL is the log likelihood) and the deviation from ideal geometry against refinement cycle. zBOND and zANGL are the Z-scores of the deviations of bond lengths and angles from ideal values: at high resolution these should approach 1.0 while at low resolution, where structures are tightly restrained to ideal values, they should approach 0.0.
[edit] TLS Refinement
TLS refinement allows you to model anisotropic displacements (e.g. motions) of the atoms of your model at medium to low resolution. It does this by constraining the allowed displacements to a rigid body model, which requires 20 free parameters per rigid group or "TLS group". To initiate TLS refinement, select "TLS & restrained refinement" from the pull-down menu at the top of the task interface.
Refmac needs to know your choice of TLS groups and the initial sets of TLS parameters. By default, Refmac will assign one TLS group per protein molecule and start with all TLS parameters set at zero. This is usually a good first choice, and can be used as a reference against which to judge more detailed TLS models. To accept the default, leave the "TLS in (optional)" selection blank. You will still need to assign a name for the "TLS out" file. As well the TLS parameters, Refmac will also refine "residual" B factors which give an atomic isotropic contribution in addition to the TLS contribution. By default, the starting values are taken from the "PDB in" file. However, it is often a good idea to reset these to a constant value before doing TLS refinement for the first time, and this can be done by selecting "Set initial Bfactors ..." in the TLS Parameters folder. All other Refmac parameters are as for restrained refinement.
Select 'Run now', and the program will run. It will perform several rounds of TLS refinement (default 10) followed by rounds of normal restrained refinement.
If you wish to define a different set of TLS groups and/or give initial values for the TLS parameters, then you need to create a TLSIN file and specify it on the "TLS in (optional)" line. This is a simple text file, and can be created by hand, although it is probably safer to use the Create/Edit TLS File task. For subsequent rounds of TLS refinement, TLSIN can be taken as the TLSOUT of the previous round.
[edit] Advanced options
[edit] Refinement mode
At the top of the task window are the options controlling refinement mode. In addition to selecting the data to refine against, you can chose what type of refinement calculation to perform. The options are:
- Review restraints. No refinement is performed, but the geometry of the model is checked.
- Restrained refinement. This is the default: the model is refined to give the best fit to the experimental density while maintaining good geometry.
- Unrestrained refinement. Do not restrain the model geometry. Only useful at high resolution.
- Rigid body refinement. Don't move individual atoms, move the whole molecular as a rigid unit. Useful at low resolutions, for example after molecular replacement.
- Structure idealisation. Don't use experimental data, just correct any errors in the geometry. (Not normally used).
- TLS and restrained refinement. Like restrained refinement, but rigid body thermal motion of the molecule is also modelled, giving a first approximation to anisotropic refinement, even at low resolution.
[edit] Refinement parameters
The follow parameters may be set:
- Number of cycles. How many steps of refinement to perform. The default of 10 is generally sufficient. If on the final cycle the Free-R factor is still decreasing, then run more cycles.
- Use hydrogen atoms. ASK GARIB
- Resolution range. Normally you should use all of your data, however if you know there are problems with the lowest or highest resolution data you can exclude them here.
- Use automatic weighting between X-ray and geometry terms. If this is unchecked, you get the possibility to set the weighting manually, and this may be useful towards the end of refinement to optimise the weighting. Lower values can be used to tighten the geometry. The value set by the automatic weighting scheme is printed in the log file.
- Use experimental sigmas to weight X-ray terms. ASK GARIB
- Refine with ... temperature factors. Isotropic temperature factors (and TLS) should be used at lower resolutions (worse than 2A). Anisotropic temperature factors may be used at higher resolutions. Mixed allows anisotropic parameters to be used for some atoms, e.g. heavy atoms.
- Exclude data with free-R level. The default of FreeR_flag = 0 is consistent with other CCP4 software.
[edit] Setup Non-Crystallographic Symmetry (NCS) Restraints
If you have more than one copy of your molecule in the asymmetric unit, and unless you have high resolution, you should probably add NCS restraints.
[edit] Scaling
By default, Refmac employs a simple scaling function. In addition, it generates a solvent mask and calculates an explicit contribution to Fcalc from the bulk (unmodelled) solvent regions. These choices are normally sufficient, but you can also choose Babinet scaling.
[edit] Geometric parameters
This folder allows you to fine tune the geometric restraints. Usually the defaults can be accepted but the following may be worth looking at:
- Bfactor. This sets the overall weighting of the B factor restraints between 1-2 and 1-3 bonded atoms, and can be optimised with respect to LLfree using the procedure of Ian Tickle http://scripts.iucr.org/cgi-bin/paper?gx5119
[edit] Related pages
Advanced refinement with Refmac5 details more advanced usage.
[edit] Program documentation
The latest version of the documentation is available from The Refmac5 manual page. This provides information on program keywords which may be used from the command line.
This page describes Refmac_5.4.0067 (CCP4 version 6.1.0).
--Kevin Cowtan 05:59, 18 April 2008 (CDT)
Updated Martyn Winn 19 November 2009
