This usually leads to a mad
chase through all possible space groups, twinning refinements, etc. and,
in my experience, often results in a lot of time being spent for no
significant improvements.
If one has a model to refine, it is actually very straightforward to
test all possible merohedral twin laws for a given unit cell and space
group. phenix has a 'jiffy' called mmtbx.twin_map_utils (in the cctbx
actually) that carries out anisotropic and bulk solvent scaling together
with the refinement of the twin fraction for a fixed model. This can
give a very quick indication which twin law results in the best twin
fraction if you have a model.
If your space group is too high (and doesn't have any twin laws) it is a
trivial matter to expand the data to a lower point group and try that.
Currently not automated in phenix, but easy to do by hand with some
cctbx tools. The same goes for MR, expanding to a lower symmetry and
trying that is straightforward and relatively easy to automate.
Sometimes however, people get confused. I once saw a case where a person
wasn't able to a beautiful Se-SAD structure with a certain software
package and decided that the space group was too low. Merging the data
in a lower SG did allow that particular software to find the sites and
phase it. The symmetry was however too low. Using different resolution
cut offs or other packages did result in a successful structure
solution. The wrong symmetry structure is still in the PDB unless it has
been redeposited.
In the end, I guess one should try everything possible and likely,
starting with the most plausible and easy solution. Automation makes a
lot of things easy (and sometimes even fast) so in the future we might
not even have to bother about the plausible bit.
HTH
Peter
- Re: [ccp4bb] how many stuck datasets are actually twinned? Peter Zwart
-