Everything said is true, but one of the most important factors in
calculating structure factors and hence Rvalues is the scaling and
solvent model. All of these are pretty inadequate - probably all protein
"crystals" have large volumes of multiply ordered atoms - water
networks, alternate conformations etc - and our calculations try to take
these into account in different ways. Hence there can be quite big
differences between R factors although the Fcalcs are very similar.
Eleanor
. Dirk Kostrewa wrote:
... yes, and this is the reason why sfcheck should be replaced by a
more modern program at the Protein Data Bank!
Best regards,
Dirk.
Am 13.07.10 22:15, schrieb Ethan Merritt:
Should be in an FAQ somewhere:
Q: Why does sfcheck not reproduce my original R factors?
A: Because instead of using the Fc in your file, sfcheck tries to
re-calculate Fc using only your atom names, coordinates and isotropic
B factors. This is bad, because it ignores any contributions to your
original Fc from things like F_partial, Anisotropic corrections,
TLS models, scattering factor corrections, riding hydrogens, etc.
My advice is to use sfcheck only to evaluate the Fobs file resulting from
your data collection. It is not a suitable tool for validating a refined
model.
The same happened when I used this final model as starting model for a
refinement with Refmac5. Again I got an Rfactor of 20.8%.
Probably you failed to describe the full model to refmac.
Refmac and phenix.refine can both handle all the contributing factors
listed above, but it may take some work to pick a the correct
corresponding set of options.
Ethan
As far as I know, Phenix uses different algorithms for the refinement
than other programs, which in some cases can make Phenix gets better
Rfactors. Could these differences be the reason for this large
difference in the calculated Rfactors? Or I have to recheck my procedure
for mistakes?
Thanks a lot in advanced.
Looking forward to hearing form you.
Cheers,
Ariel
