Dear Petr, An Rfree of 33% does happen, especially at low resolution. However my experience is that with such high Rfrees there usually is some problem. The problem might be disorder in the protein or the crystal. In this case there is not much you can do. However, before giving up, I would at least check the following (if you have not already done it):
1) run phaser with "sgalternative all". Although the celldimensions are the same, the spacegroup maybe different and even with the same space group and cell dimensions, the packing may be different. I am currently working on a project were different crystals, grown from the same protein and with the same cell dimensions and space group, have a different packing. In one case, the molecules are 90° rotated with the respect to the reference, in other cases the molecules are tilted by 20-30°. The latter structure (in the wrong orientation) could be refined to an Rfree of ~35%! 2) check for twinning. Your 4-fold symmetry might be broken to become 2-fold, and it might be in a different direction in different parts of the crystal. Good luck! Herman -----Original Message----- From: CCP4 bulletin board [mailto:ccp...@jiscmail.ac.uk] On Behalf Of Petr Kolenko Sent: Friday, December 10, 2010 1:55 PM To: CCP4BB@JISCMAIL.AC.UK Subject: Re: [ccp4bb] Space group vs. gap between Rwork and Rfree? Some refinement of my question: I am working on mutant variant (point mutation, not on the interface) of one protein already deposited in space group I432. Screw axes are excluded, NCS operator is equivalent to missing symmetry operation. To Fred, I always process data in P1 first. After that and all analysis, I completely process the data again in suggested space groups in different directories to avoid any mixture of the files. I do not know, maybe the problem is only in refinement, but I can not make the gap between Rfree and Rwork somehow reasonably stable. Many thanks to all of you for your suggestions and interest. Petr 2010/12/10 Vellieux Frederic <frederic.velli...@ibs.fr>: > Hi Petr, > > Usually IDXREF suggests more than one Bravais lattice that is > consistent with your diffraction images; hence it is (sometimes) > worthwhile trying to INTEGRATE in all possible Bravais lattices and > this allows you to eliminate a number of possibilities (poor profiles > during integration, instability of refinement, poor Rmerge values at > the level of CORRECT). This leaves you with a smaller number of > choices for the Bravais lattive. And I personally (at the end) > integrate in the correct Bravais lattice instead of integrating in P1 > and then doing the space group assignment at the level of CORRECT. I > get improved Rmerge values then. > > I've had a case where one of the students here was wrongly advising my > student about an orthorhombic space group, and my advice to carry out > integration in all Bravais lattices suggested by IDXREF, integration > led in orthorhombic gave a molecular replacement solution that did not > refine properly and a molecular replacement solution that refined in a > hexagonal space group. The Rmerge values (at the level of CORRECT) > vere very similar both for P222 and for P6 integration. > > HTH, > > Fred. > > Petr Kolenko wrote: >> >> Dear colleagues, >> >> I appreciate any help, or any suggestion with my difficult data. Many >> thanks at least for consideration. >> I work with datasets at 3.6AA of resolution. Integrated with XDS, >> scaled with SCALA. After integration and scaling in P1, POINTLESS >> suggested space group I432: >> >> Space group confidence: 0.95 >> Laue group confidence: 1.000 >> Total probability: 0.97 >> >> >> After any longer refinement (more than 20 cycles), there was always a >> big gap between Rwork and Rfree (0.22 and 0.33), although the >> structure looks quite good at this resolution. I tried also space >> group I23, here are my statistics: >> >> Data processing (I23 vs I432): >> Rmerge - 0.124 vs 0.134 >> high resolution Rmerge - 0.649 vs 0.732 low resolution Rmerge - 0.048 >> vs 0.037 >> >> The gap between Rwork and Rfree was stabilized in I23 using tight NCS >> restraints: >> >> Rwork vs Rfree: >> I23 withouth NCS: 0.228 vs 0.334 >> I23 with NCS: 0.249 vs 0.296 >> I43 : 0.228 vs 0.326 >> >> My question is, what would you recommend me to close the gap? Or is >> this Rfree pointing out a lower real symmetry (I23) than suggested by >> POINTLESS or PHENIX.XTRIAGE (I432)? I have tried different selection >> of free reflections in I432 dataset already, but with almost the same >> results. >> >> Many thanks for any response. >> Petr >> >> > -- Petr Kolenko petr.kole...@biochemtech.uni-halle.de http://kolda.webz.cz
