Hi all, I am getting quite low rmsd values for my structure. I am finishing the refinement and I have rmsd of 0.003 Å for bonds and 0.67 for angles, with R/Rfree of 0.148/0.176. The resolution is 2.0 Å. I am refining with riding hydrogens and using the automatic weight optimization. I have also tried to change the weights manually, but the more eye-pleasing rmsd values always result in worse Rfree. Is there any reason to worry about the low rmsd values and should I always use the optimized weights in refinement? I am using the latest version of Phenix. Regards, Teemu -- Teemu Haikarainen (M.Sc.) Turku Centre for Biotechnology Tykistokatu 6 BioCity, Turku FIN-20521, Finland
Hi Teemu, the statistics you quoted looks good to me. Weight optimization tries an array of different weights and chooses the one that brings lower Rfree and not yet distorted geometry (so it has upper limit on geometry deviations but doesn't have a lower limit). What pleases your eye does not necessarily please the Rfree -:) What are the Rwork/Rfree if you choose such a weight that will bring you the rmsds you like? What are the values if you do not optimize the weights? All the best! Pavel. On 3/23/11 3:01 AM, Teemu Haikarainen wrote:
Hi all,
I am getting quite low rmsd values for my structure. I am finishing the refinement and I have rmsd of 0.003 Å for bonds and 0.67 for angles, with R/Rfree of 0.148/0.176. The resolution is 2.0 Å. I am refining with riding hydrogens and using the automatic weight optimization.
I have also tried to change the weights manually, but the more “eye-pleasing” rmsd values always result in worse Rfree.
Is there any reason to worry about the low rmsd values and should I always use the optimized weights in refinement?
I am using the latest version of Phenix.
Regards, Teemu
-- Teemu Haikarainen (M.Sc.) Turku Centre for Biotechnology Tykistokatu 6 BioCity, Turku FIN-20521, Finland
I have actually played with the weights quite a bit. For example, by using the
default weights without optimization, I get the conventional rmsd values
(0.015 for bonds and 1.4 for angles) with R/Rfree of 0.123/0.189.
I guess the manual method of weight picking is not as objective as the Rfree
minimization... That is one of the reasons I would like to know if there is any
reason not to use weight optimization during refinement, whatever rmsds it
might result to.
Thanks,
Teemu
Quoting Pavel Afonine
Hi Teemu,
the statistics you quoted looks good to me. Weight optimization tries an array of different weights and chooses the one that brings lower Rfree and not yet distorted geometry (so it has upper limit on geometry deviations but doesn't have a lower limit).
What pleases your eye does not necessarily please the Rfree -:)
What are the Rwork/Rfree if you choose such a weight that will bring you the rmsds you like? What are the values if you do not optimize the weights?
All the best! Pavel.
On 3/23/11 3:01 AM, Teemu Haikarainen wrote:
Hi all,
I am getting quite low rmsd values for my structure. I am finishing the refinement and I have rmsd of 0.003 Å for bonds and 0.67 for angles, with R/Rfree of 0.148/0.176. The resolution is 2.0 Å. I am refining with riding hydrogens and using the automatic weight optimization.
I have also tried to change the weights manually, but the more eye-pleasing rmsd values always result in worse Rfree.
Is there any reason to worry about the low rmsd values and should I always use the optimized weights in refinement?
I am using the latest version of Phenix.
Regards, Teemu
-- Teemu Haikarainen (M.Sc.) Turku Centre for Biotechnology Tykistokatu 6 BioCity, Turku FIN-20521, Finland
On Wed, Mar 23, 2011 at 5:50 AM,
I guess the manual method of weight picking is not as objective as the Rfree minimization... That is one of the reasons I would like to know if there is any reason not to use weight optimization during refinement, whatever rmsds it might result to.
Yes: it takes much longer to run. I agree with Pavel, these statistics are excellent for 2.0A resolution - I wouldn't worry about the unusually low geometry deviations if everything else makes sense. -Nat
On Wed, 2011-03-23 at 12:01 +0200, Teemu Haikarainen wrote:
I have rmsd of 0.003 Å for bonds and 0.67 for angles, with R/Rfree of 0.148/0.176.
The "right" value of the bonds rmsd is a matter of debate (vigorous at times), but 0.003A to me indicates a bit of over-restraining. You may want to take a closer look at how shallow the Rfree dependence on restraint weights is (this info should be easy to extract from log-files). Maybe you don't need to go all the way to 0.003 to get most of the Rfree drop. Just to expand on this a bit, I think that the rmsd_bonds resolution dependence is really an artifact of low resolution refinement. Covalent bonds are not suddenly getting stronger when the protein is in a more disordered lattice, so one could expect that the bond length variation will be approximately the same across (almost) all proteins. The best way to estimate this "natural" level of bond rmsd is from ultrahigh resolution structures refined without restraints, and if I remember correctly, such analysis suggests something around 0.015-0.02 A. At high resolution restraints can be made sufficiently weak to (hopefully) capture the variation in bond lengths, but at lower resolution this results in the refinement of noise, correspondingly increasing the Rfree. -- "I'd jump in myself, if I weren't so good at whistling." Julian, King of Lemurs
I have rmsd of 0.003 Å for bonds and 0.67 for angles, with R/Rfree of 0.148/0.176. The "right" value of the bonds rmsd is a matter of debate (vigorous at times), but 0.003A to me indicates a bit of over-restraining. You may want to take a closer look at how shallow the Rfree dependence on restraint weights is (this info should be easy to extract from log-files). Maybe you don't need to go all the way to 0.003 to get most of the Rfree drop.
yes, we should probably make the weight optimization procedure to account for this... It's been a while since reviwed that code last time. Pavel.
participants (5)
-
Ed Pozharski -
Nathaniel Echols -
Pavel Afonine -
Teemu Haikarainen -
thaikara＠btk.fi