Also, if you have translational NCS then recent versions of Phaser can correct for the statistical effects and give you <I^2>/<I>^2 moment tests that are diagnostic of twinning. This works pretty well for 2-fold tNCS (i.e. one major Patterson peak corresponding to one or more pairs of molecules separated by the same translation). If there's higher order tNCS, then this works less well in the current version. We give some examples in the paper describing the algorithm: http://journals.iucr.org/d/issues/2013/02/00/dz5268/dz5268.pdf.
Best wishes, Randy Read On 29 Jan 2014, at 13:30, Eleanor Dodson <eleanor.dod...@york.ac.uk> wrote: > Dont forget that with twinning in apparent point group PG6/mmm the > true SG may be P6i or P3i21 > See the twinning notes: http://www.ccp4.ac.uk/dist/html/twinning.html > > > Detecting twinning can be problematic - > > My rule of thumb, following the procedure od ctruncate:: > > 0) Check the matthews coefficient for likely number of molecules. > Half a molecule must mean you are assigning too high a symmetry count. > Lots of molecules means you need to check for non-crystallographic > translation etc. > > > 1) Look at the <I^2>/<I>^2 plot after correction for anisotropy > If it isnt reasonably straight with resolution you probably have some > data problems, and these can make all the tests pretty useless. > > 2) Is there a NC translation - truncate tells you that. > If not, and the data is OK, you are unlikely to have twinning if > <I^2>/<I>^2 for acentrics is ~ 2, and the L test looks OK. > H test and Britten tests a bit more influenced by other NC symmetry > considerations > > 3) If there IS NC translation <I^2>/<I>^2 for acentrics will probably > be > 2 but the L test is still pretty reliable. > > Good luck Eleanor > > experimental phasing is tricky with perfect twinning but it has been > done. Sorry I have forgotten reference though.. > Eleanor > > > > On 29 January 2014 09:17, Kay Diederichs <kay.diederi...@uni-konstanz.de> > wrote: >> Dear Bert, >> >> as Dirk has pointed out, if P622 is the correct space group, then the >> twinning statistics printed out if you process in P6 are meaningless. >> >> Intensity statistics, like the ratio of <I^2>/<I>^2 , can be misleading if >> there is (e.g. pseudo-translational) NCS in the crystal; however, the effect >> of NCS on the value of the ratio of <I^2>/<I>^2 is opposite to that of >> twinning. Thus if a crystal is twinned and has NCS, you might not notice any >> problem in the ratio of <I^2>/<I>^2 . >> >> The other statistics, like Britton and H-test, present the intensity >> statistics in a different way, but from my understanding do not give >> substantially different information. >> >> The L-test does look at a different kind of information and therefore gives >> additional insight. >> >> If your measurements suffer from high background, diffuse scatter, ice >> rings, smeared reflections, additional crystals in the beam, or any other >> pathology, then all these tests may give distorted answers. In other words, >> even if twinning is not really present, any test designed to convert the >> deviation of data from ideality into an estimate of the twinning fraction >> will give you an alpha > 0. So my experience is: if your data are very good, >> then the tests give good answers; if the data are mediocre or bad, don't >> necessarily believe the numbers. >> >> Finally, it's not only twinning of P6 that would give you P622, it's also >> twinning of P3x21, P3x12 that gives P6y22. >> >> Hope this helps, >> >> Kay >> >> >> >> >> On Tue, 28 Jan 2014 17:26:23 +0000, Bert Van-Den-Berg >> <bert.van-den-b...@newcastle.ac.uk> wrote: >> >>> Dear all, >>> >>> I recently collected several datasets for a protein that needs experimental >>> phasing. >>> The crystals are hexagonal plates, and (automatic) data processing suggests >>> with high confidence that the space group is P622. This is where the fun >>> begins. >>> For some datasets (processed in P622), the intensity distributions are >>> normal, and the L-test (aimless, xtriage) and Z-scores (xtriage) suggest >>> that there is no twinning (twinning fractions < 0.05). However, for other >>> datasets (same cell dimensions), the intensity distributions are not normal >>> (eg Z-scores > 10). Given that twinning is not possible in P622, this >>> suggests to me that the real space group could be P6 with (near) perfect >>> twinning. >>> >>> If I now process the "normal L-test P622" datasets in P6, the twin-law >>> based tests (britton and H-test in xtriage) give high twin fractions (0.45- >>> 0.5), suggesting all my data is twinned. >>> Does this make sense (ie can one have twinning with "normal" intensity >>> distributions)? >>> If it does, would the "normal L-test" datasets have a higher probability of >>> being solvable? >>> >>> Is there any strategy for experimental phasing of (near) perfect twins? SAD >>> would be more suitable than SIR/MIR? (I also have potential heavy atom >>> derivatives). >>> >>> Thanks for any insights! >>> >>> Bert >>> ------ 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 Road E-mail: rj...@cam.ac.uk Cambridge CB2 0XY, U.K. www-structmed.cimr.cam.ac.uk
