Re: [ccp4bb] [ccp4bb] Molecular Replacement using low sequence identity templates
I guess my experience is out of date - please ignore comments on LLG! Eleanor On 21 October 2013 15:40, Randy Read wrote: > Hi Eleanor, > > Yes, the initial LLG scores in Phaser are highly dependent on the assigned > sequence identity, which is translated into an initial estimate of the > effective RMSD of the model. However, the latest versions of Phaser refine > the RMSD estimate at the end of the job and, assuming that two runs find the > same solution and the refinement manages to converge (which it usually does > these days), the LLG at the end should be pretty reproducible regardless of > the assigned sequence identity. > > Best wishes, > > Randy > > On 21 Oct 2013, at 14:42, Eleanor Dodson wrote: > >> I dont know about LLG scores - they seem extremely variable depending >> on the degree of sequence similarity you assign. >> However when you get an R/Rfree of 40%/47% that is a pretty good sign >> that at least something is correct. >> >> It isnt clear whether that is after you have placed 2 copies of the >> second component? >> Anyway - maybe try again with the refined component 2 and look for >> component 1 again? >> Eleanor >> >> >> >> On 18 October 2013 15:51, wrote: >>> Dear Jan, >>> >>> There are a few things a would do in this case. The first is to check the >>> processing to make sure the space group is really C2 and, although >>> unlikely, not some other space group. >>> >>> The second thing would be to try to place the first component. From your >>> email it is not clear to me whether or not you were able to place the first >>> component after the second component had been placed. In your case, I would >>> give both components to phaser and ask phaser to first place component 2 >>> and then component 1. >>> >>> It might be that the correct solution gets rejected because of clashes, so >>> I would also try to trim the first component, or to increase the number of >>> allowed clashes in Phaser. Although you expect two copies of your >>> heterodimer, you may have a crystal with a high solvent content and only >>> one dimer in the asymmetric unit. In this case the crystal packing should >>> make sense i.e. continuous crystal contacts in all three dimensions. >>> >>> Best, >>> Herman >>> >>> >>> -Ursprüngliche Nachricht- >>> Von: CCP4 bulletin board [mailto:CCP4BB@JISCMAIL.AC.UK] Im Auftrag von Jan >>> Félix >>> Gesendet: Freitag, 18. Oktober 2013 13:17 >>> An: CCP4BB@JISCMAIL.AC.UK >>> Betreff: [ccp4bb] Molecular Replacement using low sequence identity >>> templates >>> >>> Dear all, >>> I have a question regarding Molecular Replacement using low sequence >>> identity templates. >>> >>> I have a 2.7 Angstrom dataset of a heterodimeric protein-protein complex >>> (space group C 1 2 1, no twinning detected using xtriage). For the first >>> component homologs are available, but for the other the best found template >>> only has 20 % sequence similarity. >>> Strangely, I cannot place the first component directly, but the second >>> component can be placed (after trimming the template with chainsaw) using >>> phaser with a TFZ score of 12 and a LLG of about 1200. Although at least 2 >>> NCS copies of the heterodimer are expected based on the unit cell >>> parameters, only 1 copy of the second component gets placed. >>> If I try to place the first component based on the .sol file of the first >>> MR solution, the TFZ score for the second placement is only about 3.5, but >>> if I then try to place this second MR solution (2 >>> components) as a whole I get a RFZ of 8, TFZ of 16 and a LLG of about 1000. >>> >>> However, none of the MR solutions I obtained seems to refine in PHENIX. >>> Using autobuild does not lower the R/Rfree values which seem to get stuck >>> at an R/Rfree of 0.40/0.47. I have tried trimming off loops, simulated >>> annealing, DEN-refinement, morphing and MR-rosetta, but nothing seems to >>> improve the model.. >>> Also, in every MR solution only half of the asymmetric unit seems to be >>> filled, but phaser fails to place more units..As I am seriously starting to >>> doubt the actual content of the crystals, I tried Nearest Cell to search >>> for similar space group, but without any hits. >>> >>> So here is my question. Is it possible to get TFZ/LLG values this high in >>> C 1 2 1 with a completely incorrect model by chance, or can I assume that >>> this MR solution points out that what I think is in the crystal is actually >>> there? >>> And secondly, as I am a bit stuck here, are there any new strategies I can >>> try to tackle this problem? >>> Off course, experimental phasing is an option, but the crystals grew slowly >>> over e few months and I only had 1 drop with 1 crystal, so reproducing the >>> crystals might be though.. >>> >>> Thanks for any tips and best regards, >>> >>> Jan > > -- > Randy J. Read > Department of Haematology, University of Cambridge > Cambridge Institute for Medical Research Tel: + 44 12
Re: [ccp4bb] [ccp4bb] Molecular Replacement using low sequence identity templates
Hi Eleanor, Yes, the initial LLG scores in Phaser are highly dependent on the assigned sequence identity, which is translated into an initial estimate of the effective RMSD of the model. However, the latest versions of Phaser refine the RMSD estimate at the end of the job and, assuming that two runs find the same solution and the refinement manages to converge (which it usually does these days), the LLG at the end should be pretty reproducible regardless of the assigned sequence identity. Best wishes, Randy On 21 Oct 2013, at 14:42, Eleanor Dodson wrote: > I dont know about LLG scores - they seem extremely variable depending > on the degree of sequence similarity you assign. > However when you get an R/Rfree of 40%/47% that is a pretty good sign > that at least something is correct. > > It isnt clear whether that is after you have placed 2 copies of the > second component? > Anyway - maybe try again with the refined component 2 and look for > component 1 again? > Eleanor > > > > On 18 October 2013 15:51, wrote: >> Dear Jan, >> >> There are a few things a would do in this case. The first is to check the >> processing to make sure the space group is really C2 and, although unlikely, >> not some other space group. >> >> The second thing would be to try to place the first component. From your >> email it is not clear to me whether or not you were able to place the first >> component after the second component had been placed. In your case, I would >> give both components to phaser and ask phaser to first place component 2 and >> then component 1. >> >> It might be that the correct solution gets rejected because of clashes, so I >> would also try to trim the first component, or to increase the number of >> allowed clashes in Phaser. Although you expect two copies of your >> heterodimer, you may have a crystal with a high solvent content and only >> one dimer in the asymmetric unit. In this case the crystal packing should >> make sense i.e. continuous crystal contacts in all three dimensions. >> >> Best, >> Herman >> >> >> -Ursprüngliche Nachricht- >> Von: CCP4 bulletin board [mailto:CCP4BB@JISCMAIL.AC.UK] Im Auftrag von Jan >> Félix >> Gesendet: Freitag, 18. Oktober 2013 13:17 >> An: CCP4BB@JISCMAIL.AC.UK >> Betreff: [ccp4bb] Molecular Replacement using low sequence identity templates >> >> Dear all, >> I have a question regarding Molecular Replacement using low sequence >> identity templates. >> >> I have a 2.7 Angstrom dataset of a heterodimeric protein-protein complex >> (space group C 1 2 1, no twinning detected using xtriage). For the first >> component homologs are available, but for the other the best found template >> only has 20 % sequence similarity. >> Strangely, I cannot place the first component directly, but the second >> component can be placed (after trimming the template with chainsaw) using >> phaser with a TFZ score of 12 and a LLG of about 1200. Although at least 2 >> NCS copies of the heterodimer are expected based on the unit cell >> parameters, only 1 copy of the second component gets placed. >> If I try to place the first component based on the .sol file of the first MR >> solution, the TFZ score for the second placement is only about 3.5, but if I >> then try to place this second MR solution (2 >> components) as a whole I get a RFZ of 8, TFZ of 16 and a LLG of about 1000. >> >> However, none of the MR solutions I obtained seems to refine in PHENIX. >> Using autobuild does not lower the R/Rfree values which seem to get stuck at >> an R/Rfree of 0.40/0.47. I have tried trimming off loops, simulated >> annealing, DEN-refinement, morphing and MR-rosetta, but nothing seems to >> improve the model.. >> Also, in every MR solution only half of the asymmetric unit seems to be >> filled, but phaser fails to place more units..As I am seriously starting to >> doubt the actual content of the crystals, I tried Nearest Cell to search >> for similar space group, but without any hits. >> >> So here is my question. Is it possible to get TFZ/LLG values this high in C >> 1 2 1 with a completely incorrect model by chance, or can I assume that this >> MR solution points out that what I think is in the crystal is actually there? >> And secondly, as I am a bit stuck here, are there any new strategies I can >> try to tackle this problem? >> Off course, experimental phasing is an option, but the crystals grew slowly >> over e few months and I only had 1 drop with 1 crystal, so reproducing the >> crystals might be though.. >> >> Thanks for any tips and best regards, >> >> Jan -- Randy J. Read Department of Haematology, University of Cambridge Cambridge Institute for Medical Research Tel: + 44 1223 336500 Wellcome Trust/MRC Building Fax: + 44 1223 336827 Hills RoadE-mail: rj...@cam.ac.uk Cambridge CB2 0XY, U.K. www-structmed.cimr.cam.ac.uk
