some general refine questions
Hi folks: With the helpful advice of the phenix crew, I've re-refined 2goz, an entry in the pdb in which I used refmac5 the first time around. There was a small error in the pdb: One nucleotide (C8 in chain A) was rotated 180 degrees about the glycosylic bond from where it should have been. The net result is one oxygen atom was in the wrong place in the structure, and this was apparent within the difference Fourier. After I rotated the base and re-refined in refmac, the Rfactor and other stats remained essentially unchanged (so this fix has nothing to do with the following): My refinement results in Refmac: REMARK 3 R VALUE (WORKING SET) : 0.191 REMARK 3 FREE R VALUE : 0.241 REMARK 3 RMS DEVIATIONS FROM IDEAL VALUES COUNT RMS WEIGHT REMARK 3 BOND LENGTHS REFINED ATOMS (A): 1503 ; 0.009 ; 0.021 REMARK 3 BOND ANGLES REFINED ATOMS (DEGREES): 2331 ; 1.676 ; 3.000 Then I refined in phenix. I get REMARK Final: r_work = 0.1737 r_free = 0.2224 bonds = 0.003 angles = 0.745 So the working and free R factor are both lower, and the geometry is substantially tighter. If I relax the weights in phenix, even to get rms bonds of around 0.006, the free-R goes up, which tells me this is not something I am justified in doing. Phil Evans once told me that anything tighter than about 0.01 (conservatively) was science-fiction, in that this is within the error of the idealized libraries for bond lengths. Other people have asked me about this too, and I didn't know how to reply. Is bonds = 0.003 angles = 0.745 realistic, or have I done something wrong? Even if I do structure idealization in Refmac, I can't get the rms bonds below about 0.005 (and my understanding is that it uses the same monomer library). Why is the weight multiplied by 0.5 by default? Since changing that to 1.0 slightly increases freeR, I assume it is the default for some reason. All the best, Bill PS: Could someone point me to a quick description for depositing phenix-refined coordinates? (The latest phenix installer seems to have omitted the docs.) William G. Scott contact info: http://chemistry.ucsc.edu/~wgscott
Hi All, I have just completed the refinement of 2 structures by phenix, and I'm really really impressed. Particularly impressed by elbow and LigandFit, just wonderful. However, I would be very grateful if someone could point me to the topic already posted in the message below by Dr. Scott few days ago, exactly at: "a quick description for depositing phenix-refined coordinates?", though the full message contained very interesting points I would love to hear about (I also performed refinement by refmac/phenix, and observed kind of a similar trend for those numbers). Thanks in advance. EM *** Enrico Malito Ben-May Department for Cancer Research Center for Integrative Science, W423J The University of Chicago, 929 E. 57th Street, Chicago, IL, 60637 Tel: 773-702-4608 Fax: 773-702-6260 On Jun 30, 2007, at 1:16 PM, William Scott wrote:
Hi folks:
With the helpful advice of the phenix crew, I've re-refined 2goz, an entry in the pdb in which I used refmac5 the first time around.
There was a small error in the pdb: One nucleotide (C8 in chain A) was rotated 180 degrees about the glycosylic bond from where it should have been. The net result is one oxygen atom was in the wrong place in the structure, and this was apparent within the difference Fourier. After I rotated the base and re-refined in refmac, the Rfactor and other stats remained essentially unchanged (so this fix has nothing to do with the following):
My refinement results in Refmac:
REMARK 3 R VALUE (WORKING SET) : 0.191 REMARK 3 FREE R VALUE : 0.241
REMARK 3 RMS DEVIATIONS FROM IDEAL VALUES COUNT RMS WEIGHT REMARK 3 BOND LENGTHS REFINED ATOMS (A): 1503 ; 0.009 ; 0.021 REMARK 3 BOND ANGLES REFINED ATOMS (DEGREES): 2331 ; 1.676 ; 3.000
Then I refined in phenix.
I get
REMARK Final: r_work = 0.1737 r_free = 0.2224 bonds = 0.003 angles = 0.745
So the working and free R factor are both lower, and the geometry is substantially tighter.
If I relax the weights in phenix, even to get rms bonds of around 0.006, the free-R goes up, which tells me this is not something I am justified in doing.
Phil Evans once told me that anything tighter than about 0.01 (conservatively) was science-fiction, in that this is within the error of the idealized libraries for bond lengths.
Other people have asked me about this too, and I didn't know how to reply.
Is bonds = 0.003 angles = 0.745 realistic, or have I done something wrong? Even if I do structure idealization in Refmac, I can't get the rms bonds below about 0.005 (and my understanding is that it uses the same monomer library).
Why is the weight multiplied by 0.5 by default? Since changing that to 1.0 slightly increases freeR, I assume it is the default for some reason.
All the best,
Bill
PS: Could someone point me to a quick description for depositing phenix-refined coordinates? (The latest phenix installer seems to have omitted the docs.)
William G. Scott
contact info: http://chemistry.ucsc.edu/~wgscott
_______________________________________________ phenixbb mailing list [email protected] http://www.phenix-online.org/mailman/listinfo/phenixbb
Hi Bill and all interested in this,
Why is the weight multiplied by 0.5 by default? Since changing that to 1.0 slightly increases freeR, I assume it is the default for some reason.
the total target for refinement of coordinates in phenix.refine is defined as: Etotal = wc_scale * wxc * Exray + wc * Egeom Similar for refinement of isotropic or anisotropic ADP. Here: - Exray is least-squares (LS), maximum-likelihood (ML) or any other kind of x-ray target; - Egeom is geometry restrains target (Ebonds + Eangles + Eplanarities + Enonboned +...); - wc = 1.0 and used to exclude geometry restraints by setting wc = 0.0; - wxc ~ ratio of gradient's norms for geometry and x-ray targets as defined in (Adams et al, 1997, PNAS, Vol. 94, p. 5018); - wc_scale is 'ad hoc' scale found empirically to be ok for most of the cases at medium resolution. It should be optimized at higher and lower resolutions. The default values are most likely NOT optimal at resolutions higher then 1.5A. Currently we are working on making it resolution dependent rather than one single value for all resolutions. It is advised to play with wc_scale manually if: - Rfree or Rfree - Rwork look not realistic; - geometry deviations look not good (too tight or too loose at given resolution). However, once again, the default values for most of cases at "normal" resolution are OK. In future we are planning to work out more robust and sophisticated methodology for automatic weights determination. Pavel.
participants (3)
-
Enrico Malito -
Pavel Afonine -
William Scott