Yes - your real problem is probably that the Rfac is too low and that is
often due to have rather loose geometry or lots of unrealistic waters in
the model (RmsD on bonds of > 0.03 say..)
It is all a bit unrealistic if you say your maps are good. I suspect if
you tighten the geometry restraints by changing WEIGHT to a lower value,
or by using WEIGHT AUTO then the Rfac will go up, the Rfree will maybe
come down a little and neither the model or the maps will change in any
realistic way at all!
So you have shown (again) that Rfactors are not very reliable indicators
of correctness!
Eleanor
Santarsiero, Bernard D. wrote:
I'm sure that the relative weight between your structure factors and
geometric factors is wrong. If you use the default value, 0.3, it may or
may not be too high. In CNS, the relative weight factor is calculated, but
this is not done in REFMAC. As a rough rule-of-thumb, I aim for a rmsd of
bond lengths at 0.01A and rmsd of bond angles at 1.5deg. This works
reasonably well for both CNS and REFMAC. I think you'll find that your
rmsd is quite high for the REFMAC coordinates since you've relaxed the
geometric factors too much to match the structure factor differences.
Bernie Santarsiero
On Fri, March 9, 2007 8:22 pm, Nicholas Noinaj wrote:
Hi,
I am working on a 2.4 angstrom structure that is in its final stages of
refinement and finding that REFMAC5 is giving a large difference between R
and Rfree (10%), 0.16 and 0.29, respectively. 2Fo-Fc maps looks great and
there is very little density in the difference maps. My model covers the
entire protein sequence, except for ~5 residues on N-term and 3 residues
on C-term. I then used the model for refinement (annealing) within CNS,
and found that R and Rfree are relatively similar, 0.2 and 0.28, but in
better agreement.
I have read other posts addressing similar questions, but still unsure
what my problem could be, if there is one. While I don't mind repeating
things in CNS, I would really like to learn to be completely reliable on
CCP4 Suite as much as possible, since i really like the REFMAC5 - Coot
duo. I would appreciate any suggestions or comments you might offer.
Given how nice the density maps look and that the model passes procheck
analysis with very few suggestions for further investigations, i am unsure
when to go from here.
Thanks in advance for all feedback!
Cheers,
Nick
