Item 2: Topic discussion: NCS

SUMMARY: NCS is part of single crystal X-ray diffraction data and a useful addition to the crystallographic symmetry. All experimental data should be used for derivation of experimental results. NCS contains "error" that should be treated properly with the use of a weighting scheme. CNS and SHELXL use different approaches in structure refinement with NCS restraints.

Experts' views from outside the group:
http://www.ysbl.york.ac.uk/ccp4bb/2002/msg00083.html
http://www.ysbl.york.ac.uk/ccp4bb/2002/msg00091.html

That's quite interesting. I have not used TNT with NCS constraints or restraints. With X-plor/CNS you have both options as well. In the case of strict NCS, only one molecule is refined and then rotation/ translation matrix(es) will be applied to generate one or more copies. This does not allow any deviation between the copies. If deviations are anticipated, one should use NCS restraints and set appropriate weights. In this case, equivalent sets of atoms are superimposed and their coordinates are restrained to be close to the average position E(ncs) = SUM w(x-x(mean))**2 ( with x being the vector x,y,z). However, this cannot account for domain movements around hinges. If the molecule contains domains, NCS-related molecules might in fact display different rotations around those hinges. One can then define those regions separately. Furthermore, the backbone atoms may obey the NCS very well, whereas the side chains, especially those on the surface, usually don't. Figuring out which atoms belong to a NCS group can be a tedious job and one is sooner or later ready to compromise (i.e. give up NCS restraints partially ). In contrast, Shelxl has a better way to deal with NCS by setting up 'local NCS' restraints. The NCSY instruction selects NCS related atoms but sets up 1-4 distance restrains internally. This automatically allows for domain rotations without any problem. In my experience, it works very well and is quite flexible in that it allows you to select main chain or side chain atoms or both depending on the degree of similarity. By the way, it seems that X-plor can in fact apply local NCS similar to Shelx (unless I am mistaken), but it is much less obvious. One would have to use NOE energy terms and use distance restraints. For a crystallographer, this is a bit unfamiliar.
(Lothar Esser, NCI)

The most powerful fashion to use noncrystallographic symmetry is to reduce the number of parameters in your model with constraints. The correspondence of the separate images of the molecule is assumed to be perfect and the model does not allow any variation between copies. Basically the number of parameters in the model decreases by the number of images in the asymmetric unit. When there are real differences between the copies of the molecule, you do not want to use the noncrystallographic symmetry as constraints. To allow the data to tell you where the differences are you must allow some variation in your model. These requirements require that you incorporate the non-crystallographic symmetry as restraints. You will have to choose a weight to set the relative importance of these observations to your other observations.
(TNT Users' Guide)

Yes, I use NCS whenever it is observable. Why? NCS is in fact experimental data at either atomic (for identical copies of the molecule) or lower (for non-identical copies) resolutions, which contributes in both structure solution and refinement. For structure refinement, NCS data at atomic resolution can be used as constraints to powerfully reduce the number of refined parameters, whereas lower- resolution NCS data can be used as restraints for all or selected region(s) of the molecule with a proper weighting scheme. One way of choosing the region(s) and weights is to start the refinement with NCS restraints for the entire molecule and tighter restraints and then follow the indications of Fo-Fc.
(Xinhua Ji, NCI)