Hi Alex,
I have refined a 1.3A structure using individual anisotropic B-factors (Rfactor, 15.5%; Rfree, 17.7%). Additional use of the riding model during refinement has improvement the R values (Rf, 14.5%; Rfree, 16.6%),
this is not surprising at all -- hydrogens do contribute to the scattering.
although bond length and angle outliers did increase (anyone know why that would be?).
Different model - slightly different Xray/Restraints weights - slightly different statistics.
I have decided to deposit the model obtained using the riding model, but am unsure whether to include the hydrogen coordinates in the deposition.
If H atoms were used in refinement and final statistics calculation (such as R-factors, etc), then the H atoms have to be included into depositing model. Otherwise your statistics will not be reproducible.
The Phenix.refine documentation advises that they are left in, as they are needed to reproduce the refinement statistics. However, reviewing the high resolution structures in the PDB I found that:
Reviewing statistics in the PDB you can find many interesting things, such as Fobs labeled as Iobs (and Iobs labeled as Fobs), atoms with unknown scattering types labeled X (which makes R-factor impossible to compute), O-H distances in water longer than 10A, etc. one can go on for a long list.
1. In the case of structures refined using REFMAC or SHELX, the riding model does not seem to be included, although its use during refinement is mentioned in the pdb header.
Which is, in my opinion, not good.
2. In the case of structures refined using PHENIX, use of the riding model is sometimes indicated in the header, but hydrogen coordinates only appear to be included in the pdb in the case of structures refined to a resolution better than 1.3A.
This most likely indicates a manipulation with a file contained refined model, which is, again, in my opinion, not good.
As far as I can understand the purpose of the riding model is to provide additional restraints for refinement, so as long as its use is made evident in the pdb header (and maybe the methods section of the publication) that should be enough. Inclusion of hydrogens in the model, when you can't see them seems inappropriate.
Different programs use different libraries of ideal X-H parameters. Depending on the program, the B-factors of H atoms can be inherited from atoms they are bonded to or can be multiplied by a scale from 1 to 1.5. These all makes adding H program-dependent, and therefore statistics from such models may vary. I believe that any manipulation on the refined structure is bad. Pavel.