As with conventional B factor refinement, you can decide how many additional parameters you wish to include. In isotropic B factor refinement, you might do one overall B, or one or two per residue. With the anisotropic B's, you could likewise refine an overall anisotropic B or a TLS rigid-body displacement matrix. Another alternative I've read about but not seen widely used is to refine two structures simultaneously, allowing the two models to absorb the anisotropy. Even this would introduce only 3 or 4 extra parameters per atom, instead of the 5 extra per atom a full anisotropic refinement would yield. These methods and the full anisotropic refinement you've already done should all indicate which atoms are most in need of anisotropic parameters, although this would be empirical evidence, not the theoretical guidelines you were looking for. The big drop in Rfree does indicate you're going in the right direction, but I'd say try one of the above methods as well. If you can get as large or larger a drop in Rfree, with a more equitable drop in Rconv, you might want to stick with that parameterization.
This archive was generated by hypermail 2b29 : Tue Mar 20 2007 - 03:13:18 CST