Dear Phenix community, I am using phenix.refine version 2007_08_18_1856 to refine a 2.65A structure (with individual_sites, group_adp, TLS & NCS, wxc_scale=0.1). After refinement I am running the pdb file through the Molprobity server. Molprobity gives a clash score of 33.78 indicating the number of steric clashes (within 0.4A) per 1000 atoms. I am slightly concerned that a refinement program seems happy to allow such clashes. Is this a problem with the version of phenix.refine I am running, or is there a more obvious explanation that I have missed? Thanks, Simon
Hi, Basically I have the same observation. My structure looks good (1.7A, R=19.6 free=22.5; strategy=individual_sites+individual_adp+tls; twin fr. 0.47) with nice geometry and density, but the clashscore in Molprobity is terrible. Almost all residues form clashes. It is improved only slightly if I add hydrogens at riding positions. So I am also curious where the problem is... Karolina Dnia 11-02-2008, pon o godzinie 13:02 +0000, Simon Kolstoe napisał(a):
Dear Phenix community,
I am using phenix.refine version 2007_08_18_1856 to refine a 2.65A structure (with individual_sites, group_adp, TLS & NCS, wxc_scale=0.1). After refinement I am running the pdb file through the Molprobity server. Molprobity gives a clash score of 33.78 indicating the number of steric clashes (within 0.4A) per 1000 atoms.
I am slightly concerned that a refinement program seems happy to allow such clashes. Is this a problem with the version of phenix.refine I am running, or is there a more obvious explanation that I have missed?
Thanks,
Simon _______________________________________________ phenixbb mailing list [email protected] http://www.phenix-online.org/mailman/listinfo/phenixbb
I am slightly concerned that a refinement program seems happy to allow such clashes. Is this a problem with the version of phenix.refine I am running, or is there a more obvious explanation that I have missed?
Clash scores are not really used directly in refinement, so the refinement program doesn't really care about this. Adding hydrogens could improve the situation a bit, as would some model building do. How does your starting model look? HTH Peter
Thanks,
Simon _______________________________________________ phenixbb mailing list [email protected] http://www.phenix-online.org/mailman/listinfo/phenixbb
Are there not restraints to prevent steric clashes? In cns I seem to recall that the energy terms went through the roof if atoms got too close together. Secondly, if I remove the clashes through model building and then run phenix.refine both my R and Rfree go up, and phenix actually puts some of the clashes back again - does the community accept deposited pdb's with such obviously poor chemistry? Simon On 11 Feb 2008, at 14:51, Peter Zwart wrote:
I am slightly concerned that a refinement program seems happy to allow such clashes. Is this a problem with the version of phenix.refine I am running, or is there a more obvious explanation that I have missed?
Clash scores are not really used directly in refinement, so the refinement program doesn't really care about this. Adding hydrogens could improve the situation a bit, as would some model building do.
How does your starting model look?
HTH
Peter
Thanks,
Simon _______________________________________________ phenixbb mailing list [email protected] http://www.phenix-online.org/mailman/listinfo/phenixbb
_______________________________________________ phenixbb mailing list [email protected] http://www.phenix-online.org/mailman/listinfo/phenixbb
Simon, I saw you fixed your wxc_scale to 0.1. What happens to the clashes if you let the program refine this parameter? Carsten
-----Original Message----- From: [email protected] [mailto:[email protected]]On Behalf Of Simon Kolstoe Sent: Monday, February 11, 2008 10:01 AM To: PHENIX user mailing list Subject: Re: [phenixbb] Clashes problem
Are there not restraints to prevent steric clashes? In cns I seem to recall that the energy terms went through the roof if atoms got too close together.
Secondly, if I remove the clashes through model building and then run phenix.refine both my R and Rfree go up, and phenix actually puts some of the clashes back again - does the community accept deposited pdb's with such obviously poor chemistry?
Simon
On 11 Feb 2008, at 14:51, Peter Zwart wrote:
I am slightly concerned that a refinement program seems happy to allow such clashes. Is this a problem with the version of phenix.refine I am running, or is there a more obvious explanation that I have missed?
Clash scores are not really used directly in refinement, so the refinement program doesn't really care about this. Adding hydrogens could improve the situation a bit, as would some model building do.
How does your starting model look?
HTH
Peter
Thanks,
Simon _______________________________________________ phenixbb mailing list [email protected] http://www.phenix-online.org/mailman/listinfo/phenixbb
_______________________________________________ phenixbb mailing list [email protected] http://www.phenix-online.org/mailman/listinfo/phenixbb
_______________________________________________ phenixbb mailing list [email protected] http://www.phenix-online.org/mailman/listinfo/phenixbb
Hi, We have experienced the same problem, and what seems to work very well is the following procedure: 1) Add hydrogens to your rebuilt model with phenix.reduce: phenix.reduce input.pdb > input_h.pdb 2) Minimise the geometry of the protonated model from 1) withOUT X- ray term with: phenix.geometry_minimization --max-iterations=N input_h.pdb using a relatively low number of iterations N. You must determine this empirically by trying out a few possibilities, and chosing the highest number of iterations that improves your Molprobity clash score without over-minimising the model, as evidenced by loss of secondary structure. N=100 is a good guess. 3) Feed the geometry-minimised model to phenix.refine, as usual: phenix.refine input_h_geometry_minimized.pdb ... As Peter pointed out, this is generally an issue that one encounters at early stages of refinement, when models are still poor; therefore, in most cases you will need to mimimise geometry just once, or during the first cycles of rebuilding only. HTH, Luca ------------------------------------------------ Luca Jovine, Ph.D. Karolinska Institutet Department of Biosciences and Nutrition Hälsovägen 7, S-141 57 Huddinge, Sweden Voice: +46.(0)8.6083-301 FAX: +46.(0)8.6089-290 E-mail: [email protected] W3: http://www.biosci.ki.se/groups/ljo/ ------------------------------------------------ On 11 Feb 2008, at 16:01, Simon Kolstoe wrote:
Are there not restraints to prevent steric clashes? In cns I seem to recall that the energy terms went through the roof if atoms got too close together.
Secondly, if I remove the clashes through model building and then run phenix.refine both my R and Rfree go up, and phenix actually puts some of the clashes back again - does the community accept deposited pdb's with such obviously poor chemistry?
Simon
On 11 Feb 2008, at 14:51, Peter Zwart wrote:
I am slightly concerned that a refinement program seems happy to allow such clashes. Is this a problem with the version of phenix.refine I am running, or is there a more obvious explanation that I have missed?
Clash scores are not really used directly in refinement, so the refinement program doesn't really care about this. Adding hydrogens could improve the situation a bit, as would some model building do.
How does your starting model look?
HTH
Peter
2) Minimise the geometry of the protonated model from 1) withOUT X- ray term with:
phenix.geometry_minimization --max-iterations=N input_h.pdb
I have a question about geometry_minimization. I think wxc_scale = 0 is the same as phenix.geometry_minimization. But I have some external bond definitions defined in the eff file. So if I use wxc_scale = 0 and phenix.refine, will those external bonds be imposed on the model as the default geometries? Thanks. Jianghai
Hi Jianghai, the answers to both of your questions are YES: - running phenix.refine with "wxc_scale=0" is essentially the same as performing a pure geometry regularization (no Xray contribution is accounted for). It is just less time efficient since even if you say "wxc_scale=0", phenix.refine still compute all X-ray related things like X-ray targets, R-factors, do bulk-solvent correction etc... - yes, phenix.refine will use all your custom bond definitions regardless the value of "wxc_scale" parameter. Pavel. On 2/11/2008 12:43 PM, Jianghai Zhu wrote:
2) Minimise the geometry of the protonated model from 1) withOUT X- ray term with:
phenix.geometry_minimization --max-iterations=N input_h.pdb
I have a question about geometry_minimization. I think wxc_scale = 0 is the same as phenix.geometry_minimization. But I have some external bond definitions defined in the eff file. So if I use wxc_scale = 0 and phenix.refine, will those external bonds be imposed on the model as the default geometries?
Thanks.
Jianghai _______________________________________________ phenixbb mailing list [email protected] http://www.phenix-online.org/mailman/listinfo/phenixbb
participants (7)
-
Jianghai Zhu -
Karolina Michalska -
Luca Jovine -
Pavel Afonine -
Peter Zwart -
Schubert, Carsten [PRDUS] -
Simon Kolstoe