If your concern is vitriol against your model, then you need to ask - which one is the more plausible model?
Although there is insufficient information provided, and not considering potential complications from
twinning, your table allows some answers:
Given the absolute values of the Rs, you probably have reasonably high resolution data (1.7 A, if your
other table is right/complete). At best and unrestrained, your rms deviations could approach the variances of
small molecule structures, that, is, around 0.02 A and 2 deg. But at 1.7 A, you are still far from
a sufficient data to parameter ratio for free refinement.
So your model A is not quite plausible because the bond length variation exceeds that value.
This is consistent with a rather large R-value gap, with very low R (and perhaps with a partial gift from
twin refinement, I venture).
On the other hand, model B has an unusually tight geometry given the amount of data you likely have.
On the B-values, as noted by Eleanor, you could select sigmas of e.g. 3 5 7 9 in the 4 B-factor boxes
of refmac, this loosens the B-restraints, and observe R-free or -LLf.
For justification of the empirical 1-2 and 1-3 B-value variances, there are papers by Dale Tronrud to read.
Then you can optimize the weight, again by R-free or -LLf minimization - the Tickle papers on that
subject explain more.
As most people suggested, your optimal matrix weight is probably somewhere in between - loose
restraints but not quite unrestrained.
The optimal weights depend also to a degree on the structure itself - some are more rigid and
may keep together well by themselves, others need more guidance.
Therefore, you need to try and apply your best judgment. No recipe, sorry.
Note: In REFMAC, the actual X-ray term weight (Wa in the Xplor-Phenix-family)
is obtained as the product of the user-selectable X-ray
matrix weight times the ratio of the trace of the geometry
Hessian divided by the trace of the X-ray Hessian matrix.
The REFMAC X-ray matrix weight is therefore not the same as Wa.
Ian Tickle has kindly pointed me to the respective REFMAC
source code for verification. Garib - still true?
More questions regarding refmac auto weight will come in a separate thread.
Best, Vitri
-----Original Message-----
From: CCP4 bulletin board [mailto:[log in to unmask]] On Behalf Of Keller, Jacob
Sent: Montag, 25. Januar 2016 21:52
To: [log in to unmask]
Subject: [ccp4bb] R values Vs Geometry
Dear Crystallographers,
I seem to have an artistic situation on my hands: the decision between R and geometry.
Refinement strategy A:
R/Rfree: 0.084/0.156
RMSbonds: 0.025
RMSlengths: 2.1995
Refinement strategy B:
R/Rfree: 0.136/0.177
RMSbonds: 0.003
RMSlengths: 0.338
RSRZ scores or similar are not available in B, but there are 0 outliers in A.
Do I just flip a coin? I am inclined to go for B, since the data are twinned and are therefore somewhat less information-rich than normal, so should perhaps be guided by the geometrical book values more than normal, but the Rfree is fully 2% lower in A. Also, the R/Rfree gap is probably too big in A. Any advice/input on how to decide? Having been chastened by the recent "questionable structure" thread and the vitriol against PDB pollution, I am trying to be especially vigilant.
Jacob Keller
*******************************************
Jacob Pearson Keller, PhD
Looger Lab/HHMI Janelia Research Campus
19700 Helix Dr, Ashburn, VA 20147
email: [log in to unmask]
*******************************************
|