[ccp4bb] Stable Refinement as Low(ish) resolution - Follow up
Hi All, Thanks for all the useful comments. Buster gave me quite stable refinement and nice maps, so I've been working with that so far. I'll work through the other suggestions however, and see what gives the best results. Cheers, Rhys -- Dr Rhys Grinter Sir Henry Wellcome Fellow Monash University +61 (0)3 9902 9213 +61 (0)403 896 767
Re: [ccp4bb] Stable Refinement as Low(ish) resolution
*Ahem* *ISOLDE Tristan Croll Research Fellow Cambridge Institute for Medical Research University of Cambridge CB2 0XY > On 13 Jul 2017, at 08:02, Tristan Crollwrote: > > This is more-or-less exactly the task I'm building ISODE for. It's early > days, still quite rough around the edges and by no means complete, but if you > have a machine with a decent GPU running a modern Linux distro, then it's > easily available as a plugin to ChimeraX after installing the latest daily > build from https://www.rbvi.ucsf.edu/chimerax/download.html. Think of it as > an interactive real-space refinement environment, and you won't be far off. > This video is getting a little dated now, but should give you some idea of > what it can do: > https://drive.google.com/open?id=0B6uMjfjuw4k8aUdyTEJWeEJ1UnM. I'm currently > working on a longer, narrated demo/tutorial video showing all the latest > features (interactive position/rotamer/secondary structure assignment, a very > fun tool for shifting stretches in register, ...). > > There are some caveats: > - It doesn't yet do any building (adding or removing of atoms/residues) > - It requires all residues to be complete including hydrogens > - Currently only protein, nucleic acid and metal ions are supported (although > the metal ions need more work to be properly useful) > > Let me know if it's of interest, and we can talk more offline. > > Cheers, > > Tristan > >> On 2017-07-13 00:17, Rhys Grinter wrote: >> Dear All, >> I'm currently in the process of refining a low(ish) resolution >> structure at 3.2 Ang, with a fair level of anisotropy. I processed the >> data through the anisotropy server >> (https://services.mbi.ucla.edu/anisoscale/ [1]), which elliptically >> truncated the data to 4.0, 3.8 and 3.2 Ang. This really improved the >> maps and allowed me to trace the majority of the chain and build most >> side chains. >> The R-factors are reasonable (0.29 work and 0.35 free respectively). >> but I'm having trouble with over fitting in refinement as I continue >> to refine. What parameters/restraints would the community generally >> use when refining this kind of structure? Additionally Refmac doesn't >> seem to read the structure factors from the anisotropy server output >> file properly, giving vastly inflated R values and strange looking >> maps. >> Cheers, >> Rhys >> -- >> Dr Rhys Grinter >> Sir Henry Wellcome Fellow >> Monash University >> +61 (0)3 9902 9213 >> +61 (0)403 896 767 >> Links: >> -- >> [1] https://services.mbi.ucla.edu/anisoscale/
Re: [ccp4bb] Stable Refinement as Low(ish) resolution
This is more-or-less exactly the task I'm building ISODE for. It's early days, still quite rough around the edges and by no means complete, but if you have a machine with a decent GPU running a modern Linux distro, then it's easily available as a plugin to ChimeraX after installing the latest daily build from https://www.rbvi.ucsf.edu/chimerax/download.html. Think of it as an interactive real-space refinement environment, and you won't be far off. This video is getting a little dated now, but should give you some idea of what it can do: https://drive.google.com/open?id=0B6uMjfjuw4k8aUdyTEJWeEJ1UnM. I'm currently working on a longer, narrated demo/tutorial video showing all the latest features (interactive position/rotamer/secondary structure assignment, a very fun tool for shifting stretches in register, ...). There are some caveats: - It doesn't yet do any building (adding or removing of atoms/residues) - It requires all residues to be complete including hydrogens - Currently only protein, nucleic acid and metal ions are supported (although the metal ions need more work to be properly useful) Let me know if it's of interest, and we can talk more offline. Cheers, Tristan On 2017-07-13 00:17, Rhys Grinter wrote: Dear All, I'm currently in the process of refining a low(ish) resolution structure at 3.2 Ang, with a fair level of anisotropy. I processed the data through the anisotropy server (https://services.mbi.ucla.edu/anisoscale/ [1]), which elliptically truncated the data to 4.0, 3.8 and 3.2 Ang. This really improved the maps and allowed me to trace the majority of the chain and build most side chains. The R-factors are reasonable (0.29 work and 0.35 free respectively). but I'm having trouble with over fitting in refinement as I continue to refine. What parameters/restraints would the community generally use when refining this kind of structure? Additionally Refmac doesn't seem to read the structure factors from the anisotropy server output file properly, giving vastly inflated R values and strange looking maps. Cheers, Rhys -- Dr Rhys Grinter Sir Henry Wellcome Fellow Monash University +61 (0)3 9902 9213 +61 (0)403 896 767 Links: -- [1] https://services.mbi.ucla.edu/anisoscale/
Re: [ccp4bb] Stable Refinement as Low(ish) resolution
In addition: pdb-redo automatically makes (homology-based) hydrogen bond restraints for Refmac at this resolution. You could gibe those a try. Cheers, Robbie Sent from my Windows 10 phone From: Sudipta Bhattacharyya<mailto:sudiptabhattacharyya.iit...@gmail.com> Sent: 13 July 2017 06:25 To: CCP4BB@JISCMAIL.AC.UK<mailto:CCP4BB@JISCMAIL.AC.UK> Subject: Re: [ccp4bb] Stable Refinement as Low(ish) resolution Hi Rhys, Along with what others have suggested, you can always try prosmart refinement. I found it really helpful in my cases. Best of luck, Sudipta. Sudipta Bhattacharyya Postdoctoral research fellow University of Texas at Austin Texas, USA. On Jul 12, 2017 6:20 PM, "Rhys Grinter" <rhys.grin...@monash.edu<mailto:rhys.grin...@monash.edu>> wrote: Dear All, I'm currently in the process of refining a low(ish) resolution structure at 3.2 Ang, with a fair level of anisotropy. I processed the data through the anisotropy server (https://services.mbi.ucla.edu/anisoscale/), which elliptically truncated the data to 4.0, 3.8 and 3.2 Ang. This really improved the maps and allowed me to trace the majority of the chain and build most side chains. The R-factors are reasonable (0.29 work and 0.35 free respectively). but I'm having trouble with over fitting in refinement as I continue to refine. What parameters/restraints would the community generally use when refining this kind of structure? Additionally Refmac doesn't seem to read the structure factors from the anisotropy server output file properly, giving vastly inflated R values and strange looking maps. Cheers, Rhys -- Dr Rhys Grinter Sir Henry Wellcome Fellow Monash University +61 (0)3 9902 9213<tel:+61%203%209902%209213> +61 (0)403 896 767<tel:+61%20403%20896%20767>
Re: [ccp4bb] Stable Refinement as Low(ish) resolution
Hi Rhys, Along with what others have suggested, you can always try prosmart refinement. I found it really helpful in my cases. Best of luck, Sudipta. Sudipta Bhattacharyya Postdoctoral research fellow University of Texas at Austin Texas, USA. On Jul 12, 2017 6:20 PM, "Rhys Grinter"wrote: > Dear All, > > I'm currently in the process of refining a low(ish) resolution structure > at 3.2 Ang, with a fair level of anisotropy. I processed the data through > the anisotropy server (https://services.mbi.ucla.edu/anisoscale/), which > elliptically truncated the data to 4.0, 3.8 and 3.2 Ang. This really > improved the maps and allowed me to trace the majority of the chain and > build most side chains. > > The R-factors are reasonable (0.29 work and 0.35 free respectively). but > I'm having trouble with over fitting in refinement as I continue to refine. > What parameters/restraints would the community generally use when refining > this kind of structure? Additionally Refmac doesn't seem to read the > structure factors from the anisotropy server output file properly, giving > vastly inflated R values and strange looking maps. > > Cheers, > > Rhys > > -- > Dr Rhys Grinter > Sir Henry Wellcome Fellow > Monash University > +61 (0)3 9902 9213 <+61%203%209902%209213> > +61 (0)403 896 767 <+61%20403%20896%20767> >
Re: [ccp4bb] Stable Refinement as Low(ish) resolution
Pretty much what the others have said, but there is a low resolution refinement pipeline in CCP4 called LORESTR (LOw-REsolution STructure Refinement). Several people I know have had good results with this. It automates the process of finding homologues, generating restraints, trying different scaling parameters etc. In CCP4i it is “Run Low Resolution Refinement” on the refinement page. I don’t think it is in CCP4i2 yet. Jason. -- Dr Jason Busby Laboratory of Structural Biology School of Biological Sciences The University of Auckland Thomas Building 110 3A Symonds St Private Bag 92019 Auckland Mail Centre Auckland New Zealand ph: +64 9 3737599 ext 83888 On 13/07/2017, at 11:17 AM, Rhys Grinter> wrote: Dear All, I'm currently in the process of refining a low(ish) resolution structure at 3.2 Ang, with a fair level of anisotropy. I processed the data through the anisotropy server (https://services.mbi.ucla.edu/anisoscale/), which elliptically truncated the data to 4.0, 3.8 and 3.2 Ang. This really improved the maps and allowed me to trace the majority of the chain and build most side chains. The R-factors are reasonable (0.29 work and 0.35 free respectively). but I'm having trouble with over fitting in refinement as I continue to refine. What parameters/restraints would the community generally use when refining this kind of structure? Additionally Refmac doesn't seem to read the structure factors from the anisotropy server output file properly, giving vastly inflated R values and strange looking maps. Cheers, Rhys -- Dr Rhys Grinter Sir Henry Wellcome Fellow Monash University +61 (0)3 9902 9213 +61 (0)403 896 767
Re: [ccp4bb] Stable Refinement as Low(ish) resolution
Dear Rhys, I second Roger on the use of jelly-body refinement. In addition, give Buster a try. It sometimes does magic at lowish resolution. All best. Andreas On Thu, Jul 13, 2017 at 1:17 AM, Rhys Grinterwrote: > Dear All, > > I'm currently in the process of refining a low(ish) resolution structure > at 3.2 Ang, with a fair level of anisotropy. I processed the data through > the anisotropy server (https://services.mbi.ucla.edu/anisoscale/), which > elliptically truncated the data to 4.0, 3.8 and 3.2 Ang. This really > improved the maps and allowed me to trace the majority of the chain and > build most side chains. > > The R-factors are reasonable (0.29 work and 0.35 free respectively). but > I'm having trouble with over fitting in refinement as I continue to refine. > What parameters/restraints would the community generally use when refining > this kind of structure? Additionally Refmac doesn't seem to read the > structure factors from the anisotropy server output file properly, giving > vastly inflated R values and strange looking maps. > > Cheers, > > Rhys > > -- > Dr Rhys Grinter > Sir Henry Wellcome Fellow > Monash University > +61 (0)3 9902 9213 <+61%203%209902%209213> > +61 (0)403 896 767 <+61%20403%20896%20767> >
Re: [ccp4bb] Stable Refinement as Low(ish) resolution
Besides empirically adjusting the weighting factor for X-ray data to increase the geometric constraints, have you tried jelly-body refinement or refinement with external constraints? The latter two methods can be helpful when refining low resolution data. Roger Rowlett On Jul 12, 2017 7:33 PM, "Christine Gee"wrote: > Hi Rhys, > > In Refmac, you can change the weighting of the Xray terms vs the geometry > restraints rather than use what is automatically assigned. Have you tried > this? In Phenix you can use secondary structure restraints (probably you > can do this in Refmac too, but I haven't tried) and try checking the box > optimize stereochemistry/Xray weight. This will probably help if your over > fitting is moving things into density but making the geometry sub optimal. > > Regards > Christine > > On Wed, Jul 12, 2017 at 4:17 PM, Rhys Grinter > wrote: > >> Dear All, >> >> I'm currently in the process of refining a low(ish) resolution structure >> at 3.2 Ang, with a fair level of anisotropy. I processed the data through >> the anisotropy server (https://services.mbi.ucla.edu/anisoscale/), which >> elliptically truncated the data to 4.0, 3.8 and 3.2 Ang. This really >> improved the maps and allowed me to trace the majority of the chain and >> build most side chains. >> >> The R-factors are reasonable (0.29 work and 0.35 free respectively). but >> I'm having trouble with over fitting in refinement as I continue to refine. >> What parameters/restraints would the community generally use when refining >> this kind of structure? Additionally Refmac doesn't seem to read the >> structure factors from the anisotropy server output file properly, giving >> vastly inflated R values and strange looking maps. >> >> Cheers, >> >> Rhys >> >> -- >> Dr Rhys Grinter >> Sir Henry Wellcome Fellow >> Monash University >> +61 (0)3 9902 9213 <+61%203%209902%209213> >> +61 (0)403 896 767 <+61%20403%20896%20767> >> > >
Re: [ccp4bb] Stable Refinement as Low(ish) resolution
Hi Rhys, In Refmac, you can change the weighting of the Xray terms vs the geometry restraints rather than use what is automatically assigned. Have you tried this? In Phenix you can use secondary structure restraints (probably you can do this in Refmac too, but I haven't tried) and try checking the box optimize stereochemistry/Xray weight. This will probably help if your over fitting is moving things into density but making the geometry sub optimal. Regards Christine On Wed, Jul 12, 2017 at 4:17 PM, Rhys Grinterwrote: > Dear All, > > I'm currently in the process of refining a low(ish) resolution structure > at 3.2 Ang, with a fair level of anisotropy. I processed the data through > the anisotropy server (https://services.mbi.ucla.edu/anisoscale/), which > elliptically truncated the data to 4.0, 3.8 and 3.2 Ang. This really > improved the maps and allowed me to trace the majority of the chain and > build most side chains. > > The R-factors are reasonable (0.29 work and 0.35 free respectively). but > I'm having trouble with over fitting in refinement as I continue to refine. > What parameters/restraints would the community generally use when refining > this kind of structure? Additionally Refmac doesn't seem to read the > structure factors from the anisotropy server output file properly, giving > vastly inflated R values and strange looking maps. > > Cheers, > > Rhys > > -- > Dr Rhys Grinter > Sir Henry Wellcome Fellow > Monash University > +61 (0)3 9902 9213 <+61%203%209902%209213> > +61 (0)403 896 767 <+61%20403%20896%20767> >
[ccp4bb] Stable Refinement as Low(ish) resolution
Dear All, I'm currently in the process of refining a low(ish) resolution structure at 3.2 Ang, with a fair level of anisotropy. I processed the data through the anisotropy server (https://services.mbi.ucla.edu/anisoscale/), which elliptically truncated the data to 4.0, 3.8 and 3.2 Ang. This really improved the maps and allowed me to trace the majority of the chain and build most side chains. The R-factors are reasonable (0.29 work and 0.35 free respectively). but I'm having trouble with over fitting in refinement as I continue to refine. What parameters/restraints would the community generally use when refining this kind of structure? Additionally Refmac doesn't seem to read the structure factors from the anisotropy server output file properly, giving vastly inflated R values and strange looking maps. Cheers, Rhys -- Dr Rhys Grinter Sir Henry Wellcome Fellow Monash University +61 (0)3 9902 9213 +61 (0)403 896 767