Re: [ccp4bb] Observation:Parameter ratio
Dear Prem, you also need to count the number of restraints as observations and constraints, which reduce the number of parameters. The number of restraints at this resolution might exceed the number of observations. If I remember correctly, the numbers are listed in the refmac5 log file and probably also in the PDB file header after refinement. The difference between Rfree and R1 is a measure for overfitting, see https:// doi.org/10.1107/S0907444999016868 and https://doi.org/10.1107/ S0907444997013875 Best, Tim On Saturday, February 17, 2018 8:46:39 PM CET Prem Prakash wrote: > Dear all, > How observation is to parameter ratio is determined for a crystal > structure. Suppose If we have only one observation for each parameter at a > resolution of say 2.9 Angstrom, how terribly it is easy to overfitt for the > data to get a misleading agreement between model and experiment ? Is there > any thumbrule for this ratio, and how it is dependent on the resolution of > the data. Please shed some light on these aspects. > Thank you > With kind regards > Prem -- -- Paul Scherrer Institut Tim Gruene - persoenlich - OFLC/104 CH-5232 Villigen PSI phone: +41 (0)56 310 5297 GPG Key ID = A46BEE1A signature.asc Description: This is a digitally signed message part.
Re: [ccp4bb] Observation:parameter ratio
Dear Hugh Morgan, this ratio is only true if you refine your atomic model without any geometrical restraints against the observed X-ray data. If you include parameters for ideal geometry, you add ideal bond lengths, angles, chiral volumes, planar groups and dihedral angles as additional observations. Phenix.refine reports the number of geometrical restraints. When I look at the output of a 30 kDa protein and a 400 kDa protein, that I was recently refining, I get the following average number of observed geometrical restraints per atom: 1.0-1.1 bond lenghts / atom 1.4 bond angles / atom 0.15 chiral volumes / atom 0.17-0.18 planar groups / atom 0.37-0.38 dihedral angles / atom In total, these sum up to ~3.1-3.2 observed geometrical restraints/ atom. If you add these to the ~0.5 observed X-ray amplitudes/atom (~3.6-3.7), you get a ratio observed:parameters 1 if you refine x,y,z (3 parameters/atom), and also 1 if you refine x,y,z,B (4 parameters/atom) with B-restraints for bonded atoms (which I count as the same number of observed B-restraints as observed bond lengths). If you refine instead x,y,z,grouped B-factors with 1 group per residue, you get with 7.6-8.0 atoms per residue, approximately 0.12-0.13 parameters per atom, so 3.12-3.13 parameters/atom, and your ratio observed:parameters is also 1. I hope, this helps a little bit arguing with the referee. Best regards, Dirk. Am 23.09.2009 um 14:48 schrieb hugh morgan: Hi, We are attempting to address a referees comment concerning our data/parameter ratio for a 3.4 A structure. Although the ratio is only 0.5, the R/Rfree is 25/29, which seems acceptable. Ideally I would like to include a reference showing a graph or table of data:parameter vs resolution but I have been unable to find such a paper. Please could someone point me in the right direction. Thanks in advance Hugh P.S. It would be nice to know if this table has been published http://www-structmed.cimr.cam.ac.uk/Course/Basic_refinement/Refinement.html or something similar to this, which includes higher resolution structures http://dirac.cnrs-orleans.fr/plone/Members/hinsen/data-parameter-ratio-in-x-ray-structure-determination *** Dirk Kostrewa Gene Center, A 5.07 Ludwig-Maximilians-University Feodor-Lynen-Str. 25 81377 Munich Germany Phone: +49-89-2180-76845 Fax:+49-89-2180-76999 E-mail: kostr...@genzentrum.lmu.de WWW:www.genzentrum.lmu.de ***
Re: [ccp4bb] Observation:parameter ratio
Hi Hugh A data/parameter ratio of 0.5 does seem low even at 3.4 Ang, I would expect it closer to 1 (say 0.8 to 0.9) depending on the solvent content. Why is it low - is your completeness low? What matters from the point of view of what Rfree you should expect, all being well, is the observation/parameter ratio, i.e. counting restraints as observations. Your Rfree of 29 looks low for Rwork = 25 given that your observation/parameter ratio must be close to 1, unless of course you have increased it by adding a lot of additional restraints. Also you didn't say what are your parameters: are you refining individual B factors for example? - if so this will significantly reduce the observation/parameter ratio and is probably not advisable at 3.4 Ang. Cheers -- Ian -Original Message- From: owner-ccp...@jiscmail.ac.uk [mailto:owner-ccp...@jiscmail.ac.uk] On Behalf Of hugh morgan Sent: 23 September 2009 13:48 To: CCP4BB@JISCMAIL.AC.UK Subject: Observation:parameter ratio Hi, We are attempting to address a referees comment concerning our data/parameter ratio for a 3.4 A structure. Although the ratio is only 0.5, the R/Rfree is 25/29, which seems acceptable. Ideally I would like to include a reference showing a graph or table of data:parameter vs resolution but I have been unable to find such a paper. Please could someone point me in the right direction. Thanks in advance Hugh P.S. It would be nice to know if this table has been published http://www- structmed.cimr.cam.ac.uk/Course/Basic_refinement/Refinement.html or something similar to this, which includes higher resolution structures http://dirac.cnrs-orleans.fr/plone/Members/hinsen/data-parameter-ratio-i n- x-ray-structure-determination Disclaimer This communication is confidential and may contain privileged information intended solely for the named addressee(s). It may not be used or disclosed except for the purpose for which it has been sent. If you are not the intended recipient you must not review, use, disclose, copy, distribute or take any action in reliance upon it. If you have received this communication in error, please notify Astex Therapeutics Ltd by emailing i.tic...@astex-therapeutics.com and destroy all copies of the message and any attached documents. Astex Therapeutics Ltd monitors, controls and protects all its messaging traffic in compliance with its corporate email policy. The Company accepts no liability or responsibility for any onward transmission or use of emails and attachments having left the Astex Therapeutics domain. Unless expressly stated, opinions in this message are those of the individual sender and not of Astex Therapeutics Ltd. The recipient should check this email and any attachments for the presence of computer viruses. Astex Therapeutics Ltd accepts no liability for damage caused by any virus transmitted by this email. E-mail is susceptible to data corruption, interception, unauthorized amendment, and tampering, Astex Therapeutics Ltd only send and receive e-mails on the basis that the Company is not liable for any such alteration or any consequences thereof. Astex Therapeutics Ltd., Registered in England at 436 Cambridge Science Park, Cambridge CB4 0QA under number 3751674
Re: [ccp4bb] Observation:parameter ratio
Take a look at this: Brunger, A., DeLaBarre, B., Davies, J. Weis, W. X-ray structure determination at low resolution. ACTA CRYSTALLOGRAPHICA SECTION D-BIOLOGICAL CRYSTALLOGRAPHY 65, 128-133 (2009). The paper quotes Wayne Hendrickson (says submitted) regarding determinancy point (i.e. where the number of observations becomes equal to the number of flexible torsions (of course, in this extreme case gemoetric restraints essentially become constraints and you don't refine B-factrors at all. With 50% solvent this is ~5.4A, and you are far from that limit. It is rather surprising that any reviewer anywhere would have doubts about possibility to refine structural model at 3.4A. While I don't know what the referee said *exactly*, but he/she may be unaware of this http://www.rcsb.org/pdb/statistics/histogram.do?mdcat=refinemditem=ls_d_res_highminLabel=0maxLabel=5numOfbars=10name=Resolution On Wed, 2009-09-23 at 12:48 +, hugh morgan wrote: Hi, We are attempting to address a referees comment concerning our data/parameter ratio for a 3.4 A structure. Although the ratio is only 0.5, the R/Rfree is 25/29, which seems acceptable. Ideally I would like to include a reference showing a graph or table of data:parameter vs resolution but I have been unable to find such a paper. Please could someone point me in the right direction. Thanks in advance Hugh P.S. It would be nice to know if this table has been published http://www-structmed.cimr.cam.ac.uk/Course/Basic_refinement/Refinement.html or something similar to this, which includes higher resolution structures http://dirac.cnrs-orleans.fr/plone/Members/hinsen/data-parameter-ratio-in-x-ray-structure-determination --
Re: [ccp4bb] Observation:parameter ratio
Dear Ed Pozharski, Am 23.09.2009 um 15:48 schrieb Ed Pozharski: Take a look at this: Brunger, A., DeLaBarre, B., Davies, J. Weis, W. X-ray structure determination at low resolution. ACTA CRYSTALLOGRAPHICA SECTION D-BIOLOGICAL CRYSTALLOGRAPHY 65, 128-133 (2009). The paper quotes Wayne Hendrickson (says submitted) regarding determinancy point (i.e. where the number of observations becomes equal to the number of flexible torsions (of course, in this extreme case gemoetric restraints essentially become constraints and you don't refine B-factrors at all. With 50% solvent this is ~5.4A, and you are far from that limit. yes, but this is only in case of torsion angle refinement. For x,y,z the determinancy point is ~3 A. It is rather surprising that any reviewer anywhere would have doubts about possibility to refine structural model at 3.4A. While I don't know what the referee said *exactly*, but he/she may be unaware of this http://www.rcsb.org/pdb/statistics/histogram.do?mdcat=refinemditem=ls_d_res_highminLabel=0maxLabel=5numOfbars=10name=Resolution On Wed, 2009-09-23 at 12:48 +, hugh morgan wrote: Hi, We are attempting to address a referees comment concerning our data/parameter ratio for a 3.4 A structure. Although the ratio is only 0.5, the R/Rfree is 25/29, which seems acceptable. Ideally I would like to include a reference showing a graph or table of data:parameter vs resolution but I have been unable to find such a paper. Please could someone point me in the right direction. Thanks in advance Hugh P.S. It would be nice to know if this table has been published http://www-structmed.cimr.cam.ac.uk/Course/Basic_refinement/Refinement.html or something similar to this, which includes higher resolution structures http://dirac.cnrs-orleans.fr/plone/Members/hinsen/data-parameter-ratio-in-x-ray-structure-determination -- *** Dirk Kostrewa Gene Center, A 5.07 Ludwig-Maximilians-University Feodor-Lynen-Str. 25 81377 Munich Germany Phone: +49-89-2180-76845 Fax:+49-89-2180-76999 E-mail: kostr...@genzentrum.lmu.de WWW:www.genzentrum.lmu.de ***
Re: [ccp4bb] Observation:parameter ratio
Brunger, A., DeLaBarre, B., Davies, J. Weis, W. X-ray structure determination at low resolution. ACTA CRYSTALLOGRAPHICA SECTION D-BIOLOGICAL CRYSTALLOGRAPHY 65, 128-133 (2009). The paper quotes Wayne Hendrickson (says submitted) regarding determinancy point (i.e. where the number of observations becomes equal to the number of flexible torsions (of course, in this extreme case gemoetric restraints essentially become constraints and you don't refine B-factrors at all. With 50% solvent this is ~5.4A, and you are far from that limit. From a quick glance at that paper (although not the submitted one), it looks like this determinancy point may only apply to cases where components from high resolution structures are being used at lower resolution (independent components of a complex at high res vs the full complex at low res). This may or may not apply to cases where an initial model refined at higher resolution is not available, although I'm leaning more towards the not applicable side at the moment (my understanding was that some level of over-determination was required. Exactly what minimum level of over-determination is required is something I haven't been able to find out yet). It is rather surprising that any reviewer anywhere would have doubts about possibility to refine structural model at 3.4A. While I don't know what the referee said *exactly*, but he/she may be unaware of this http://www.rcsb.org/pdb/statistics/histogram.do?mdcat=refinemditem=ls_d_res_highminLabel=0maxLabel=5numOfbars=10name=Resolution At 3.4 Angstroms, refinement should definitely be possible. But I might not point a reviewer at that histogram. Assuming mdcat=refine is limiting results to entries that have been refined, that table has entries in the 20 to 70 Angstrom range, where I'd have doubts about the validity of any refinement done. So I'm not sure that the PDB is filtering for cases of acceptable refinement vs any refinement at all (I've seen REMARK lines in deposited entries where the authors reported that the refinement statistics were not necessarily valid given the resolution of the structure; it may be easier to do a few cycles of refinement to satisfy deposition requirements than deposit an unrefined structure at low resolution). On Wed, 2009-09-23 at 12:48 +, hugh morgan wrote: Hi, We are attempting to address a referees comment concerning our data/parameter ratio for a 3.4 A structure. Although the ratio is only 0.5, the R/Rfree is 25/29, which seems acceptable. Ideally I would like to include a reference showing a graph or table of data:parameter vs resolution but I have been unable to find such a paper. Please could someone point me in the right direction. Thanks in advance Hugh P.S. It would be nice to know if this table has been published http://www-structmed.cimr.cam.ac.uk/Course/Basic_refinement/Refinement.html or something similar to this, which includes higher resolution structures http://dirac.cnrs-orleans.fr/plone/Members/hinsen/data-parameter-ratio-in-x-ray-structure-determination
Re: [ccp4bb] Observation:parameter ratio
I've seen this estimate of the determinancy point for torsion angle refinement presented by Axel Brunger at ACA meeting this summer. I don't remember all the details, but no, it did not refer to the approach described in the paper where higher resolution model is used as a restraint. It was simply done by equalizing the number of torsions in the protein of certain molecular weight and number of reflections obtained at certain resolution assuming certain solvent content. His point (assuming that I understood it correctly) was that when you are working with 5A data your are definitely at the limit where even torsional refinement may become underdetermined. Thus you need more restraints, and it seems that you may be able to get them from higher resolution structures of, say, subunits of the same protein. You make an excellent point about the histogram - just because it's in PDB doesn't mean it's right. I've seen 2A structures there with waters that had hydrogens, but I won't condone that :) On Wed, 2009-09-23 at 10:31 -0500, Pete Meyer wrote: Brunger, A., DeLaBarre, B., Davies, J. Weis, W. X-ray structure determination at low resolution. ACTA CRYSTALLOGRAPHICA SECTION D-BIOLOGICAL CRYSTALLOGRAPHY 65, 128-133 (2009). The paper quotes Wayne Hendrickson (says submitted) regarding determinancy point (i.e. where the number of observations becomes equal to the number of flexible torsions (of course, in this extreme case gemoetric restraints essentially become constraints and you don't refine B-factrors at all. With 50% solvent this is ~5.4A, and you are far from that limit. From a quick glance at that paper (although not the submitted one), it looks like this determinancy point may only apply to cases where components from high resolution structures are being used at lower resolution (independent components of a complex at high res vs the full complex at low res). This may or may not apply to cases where an initial model refined at higher resolution is not available, although I'm leaning more towards the not applicable side at the moment (my understanding was that some level of over-determination was required. Exactly what minimum level of over-determination is required is something I haven't been able to find out yet). It is rather surprising that any reviewer anywhere would have doubts about possibility to refine structural model at 3.4A. While I don't know what the referee said *exactly*, but he/she may be unaware of this http://www.rcsb.org/pdb/statistics/histogram.do?mdcat=refinemditem=ls_d_res_highminLabel=0maxLabel=5numOfbars=10name=Resolution At 3.4 Angstroms, refinement should definitely be possible. But I might not point a reviewer at that histogram. Assuming mdcat=refine is limiting results to entries that have been refined, that table has entries in the 20 to 70 Angstrom range, where I'd have doubts about the validity of any refinement done. So I'm not sure that the PDB is filtering for cases of acceptable refinement vs any refinement at all (I've seen REMARK lines in deposited entries where the authors reported that the refinement statistics were not necessarily valid given the resolution of the structure; it may be easier to do a few cycles of refinement to satisfy deposition requirements than deposit an unrefined structure at low resolution). On Wed, 2009-09-23 at 12:48 +, hugh morgan wrote: Hi, We are attempting to address a referees comment concerning our data/parameter ratio for a 3.4 A structure. Although the ratio is only 0.5, the R/Rfree is 25/29, which seems acceptable. Ideally I would like to include a reference showing a graph or table of data:parameter vs resolution but I have been unable to find such a paper. Please could someone point me in the right direction. Thanks in advance Hugh P.S. It would be nice to know if this table has been published http://www-structmed.cimr.cam.ac.uk/Course/Basic_refinement/Refinement.html or something similar to this, which includes higher resolution structures http://dirac.cnrs-orleans.fr/plone/Members/hinsen/data-parameter-ratio-in-x-ray-structure-determination --
Re: [ccp4bb] Observation:parameter ratio
On Wed, 2009-09-23 at 16:05 +0200, Dirk Kostrewa wrote: yes, but this is only in case of torsion angle refinement. For x,y,z the determinancy point is ~3 A. True, but to count all the x,y,z,b as parameters is only sensible with unrestrained refinement. If restraints are properly imposed, the effective number of spatial coordinate parameters will be close to the number of flexible torsion angles. I don't know the exact context of the referee's comments. but it's hard to imagine that anyone would argue that structures can't be refined at 3.4A resolution. How much one can learn at such resolution is a different story. --