Hi Herman, Tighter restraints typically close the gap between R and R-free. This does not mean one should just tighten the restraints to satisfy one's own (or a referee's) idea of what the gap should be. I don't think there is a clear target of how large or small the gap should be. If you optimize the restraints to get the best (free) likelihood, you usually get a reasonable R gap without explicitly optimizing it.
Cheers, Robbie > -----Original Message----- > From: CCP4 bulletin board [mailto:CCP4BB@JISCMAIL.AC.UK] On Behalf Of > Eleanor Dodson > Sent: Friday, June 21, 2013 14:21 > To: CCP4BB@JISCMAIL.AC.UK > Subject: Re: [ccp4bb] AW: Twinning problem - almost solved. > > At your resolution that seems to me a reasonable gap between R and Rfree? > Eleanor > > On 21 Jun 2013, at 12:28, herman.schreu...@sanofi.com wrote: > > > Dear Bulletin Board, > > > > After some headbanging (Refmac5 had helpfully created gap records for all > insertions and deletions present in the structure), I got refmac5 running with > the TWIN option. Refmac5 also found the k,h,-l domain and rejected the > other possible domains because they were too small. The Rfactor's are now > extremely good: ~14% and the Rfree's are for me acceptable: ~24%. Since I > found the difference between R and Rfree somewhat large, I have been > playing with the weighting. By using a weight of 0.01, I can bring the Rfactor > up to 18%, but the Rfree stays about the same or even gets a little worse. > > > > My question: is there a way to bring R and Rfree closer together, or is it > related to the twinned data and is it something we have to live with? > > > > Best regards, > > Herman > > > > > > -----Ursprüngliche Nachricht----- > > Von: CCP4 bulletin board [mailto:CCP4BB@JISCMAIL.AC.UK] Im Auftrag von > Miller, Mitchell D. > > Gesendet: Donnerstag, 20. Juni 2013 17:43 > > An: CCP4BB@JISCMAIL.AC.UK > > Betreff: Re: [ccp4bb] Twinning problem > > > > You are welcome. Let me also for the benefit of others who may search > the archives in the future, let me correct two errors below - (typo and a miss- > recollection). > > > > Specially, I was thinking that phenix.refine was now able to refine multiple > twin laws, but according to Nat Echols on the phenix mailing list > http://phenix-online.org/pipermail/phenixbb/2013-March/019538.html > > phenix.refine only handles 1 twin law at this time. > > (My typo was that and our second structure was 3nuz with twin fractions > 0.38, 0.32, 0.16 and 0.14 -- not 2nuz). > > > > A useful search for deposited structures mentioning tetartohedral > http://www.ebi.ac.uk/pdbe- > srv/view/search?search_type=all_text&text=TETARTOHEDRALLY+OR+TETAR > TOHEDRAL > > > > Regards, > > Mitch > > > > > > -----Original Message----- > > From: CCP4 bulletin board [mailto:CCP4BB@JISCMAIL.AC.UK] On Behalf Of > herman.schreu...@sanofi.com > > Sent: Thursday, June 20, 2013 8:04 AM > > To: CCP4BB@JISCMAIL.AC.UK > > Subject: [ccp4bb] AW: Twinning problem > > > > Dear Mitch (and Philip and Phil), > > > > It is clear that I should give refmac a go with the non-detwinned F's and just > the TWIN command. > > > > Thank you for your suggestions, > > Herman > > > > > > > > -----Ursprüngliche Nachricht----- > > Von: Miller, Mitchell D. [mailto:mmil...@slac.stanford.edu] > > Gesendet: Donnerstag, 20. Juni 2013 16:18 > > An: Schreuder, Herman R&D/DE > > Betreff: RE: Twinning problem > > > > Hi Herman, > > Have you considered the possibility of your crystals being tetartohedral > twinned. That is more than one of the twin laws may apply to your crystals. > > E.g. in P32 it is possible to have tetartohedral twinning which would have > > 4 twin domains - (h,k,l), (k,h,-l), (-h,-k,l) and (-k,-h,-l). Perfect tetartohedral > twinning of P3 would merge in P622 and each twin domain would have a > faction of 0.25. > > > > We have had 2 cases like this (the first 2PRX was before there was support > for this type of twinning except for in shelxl and we ended up with refined > twin fractions of 0.38, 0.28, 0.19, 0.15 for the deposited crystal and a 2nd > crystal that we did not deposit had twin fractions of 0.25, 0.27, 0.17, 0.31). > The 2nd case we had was after support for twining (including tetartohedral > twinning) was added to refmac (and I think phenix.refine can also handle > this). For 2NUZ, it was P32 with refined twin fractions of 0.25, 0.27, 0.17, 0.31. > > > > Pietro Roversi wrote a review of tetartohedral twinning for the CCP4 > proceedings issues of acta D http://dx.doi.org/10.1107/S0907444912006737 > > > > I would try refinement with refmac using the original (non-detwinned F's) > with just the TWIN command to see if it ends up keeping twin fractions for all > 3 operators (4 domains) -- especially with crystals 1 and 3 which appear to > have the largest estimates of the other twin fractions. > > > > Regards, > > Mitch > > > > > > ========================================== > > Mitchell Miller, Ph.D. > > Joint Center for Structural Genomics > > Stanford Synchrotron Radiation Lightsource > > 2575 Sand Hill Rd -- SLAC MS 99 > > Menlo Park, CA 94025 > > Phone: 1-650-926-5036 > > FAX: 1-650-926-3292 > > > > > > -----Original Message----- > > From: CCP4 bulletin board [mailto:CCP4BB@JISCMAIL.AC.UK] On Behalf Of > herman.schreu...@sanofi.com > > Sent: Thursday, June 20, 2013 6:47 AM > > To: CCP4BB@JISCMAIL.AC.UK > > Subject: [ccp4bb] Twinning problem > > > > Dear Bulletin Board, > > > > Prodded by pdb annotators, which are very hesitant to accept coordinate > files when their Rfactor does not correspond with our Rfactor, I had a look > again into some old data sets, which I suspect are twinned. Below are the > results of some twinning tests with the Detwin program (top value: all > reflections, lower value: reflections > Nsig*obs (whatever that may mean). > The space group is P32, the resolution is 2.3 - 2.6 Å and data are reasonable > complete: 95 - 100%. > > > > From the Detwin analysis, it seems that the crystals are twinned with twin > operator k,h,-l with a twinning fraction of 0.3 for crystal 1, 0.15 for crystal 2 > and 0.4 for crystal 3. Crystal 2 can be refined while ignoring twinning to get > acceptable but not stellar R and Rfree values. However, when I try to detwin > Fobs of e.g. crystal 1 (twinning fraction 0.3), R and Rfree values stay about > the same, whatever twinning fraction I try. At the time, I used the CNS > detwin_perfect protocol to detwin using Fcalcs, which brought the Rfactors > in acceptable range, but I do not feel that was the perfect solution. Ignoring > twinning on e.g. crystal 1 produces an Rfactor of 22% and an Rfree of 29% > > > > Do you have any idea what could be going on? > > > > Thank you for your help! > > Herman > > > > > > > > Crystal 1: > > > > operator -h,-k,l > > Suggests Twinning factor (0.5-H): 0.113 > > Suggests Twinning factor (0.5-H): 0.147 > > > > operator: k,h,-l > > Suggests Twinning factor (0.5-H): 0.277 > > Suggests Twinning factor (0.5-H): 0.323 > > > > operator -k,-h,-l > > Suggests Twinning factor (0.5-H): 0.101 > > Suggests Twinning factor (0.5-H): 0.134 > > > > > > Crystal 2: > > > > operator -h,-k,l > > Suggests Twinning factor (0.5-H): 0.077 > > Suggests Twinning factor (0.5-H): 0.108 > > > > operator: k,h,-l > > Suggests Twinning factor (0.5-H): 0.126 > > Suggests Twinning factor (0.5-H): 0.161 > > > > operator -k,-h,-l > > Suggests Twinning factor (0.5-H): 0.072 > > Suggests Twinning factor (0.5-H): 0.106 > > > > > > Crystal 3: > > > > operator -h,-k,l > > Suggests Twinning factor (0.5-H): 0.123 > > Suggests Twinning factor (0.5-H): 0.149 > > > > operator: k,h,-l > > Suggests Twinning factor (0.5-H): 0.393 > > Suggests Twinning factor (0.5-H): 0.433 > > > > operator -k,-h,-l > > Suggests Twinning factor (0.5-H): 0.110 > > Suggests Twinning factor (0.5-H): 0.133 > > > > > >
