 |
 |
Dear Yury I looked at this data and such drop of R/Rfree when you switch from non-twin to twin in the presence of twinning is expected. When you said that electron density is too good I was afraid that there is strong bias. But at least for the case from the pdb (2gl5) you can see a lot of features that are not in the model (double conformations, waters, end even some unexplained ligands etc). In general, maps, especially unexplained parts are best indicators of bias towards the errors in the model. Regards Garib On 10 Feb 2011, at 22:06, Patskovsky Yury wrote: > Thanks, Garib > > It might be the case. > As a matter of fact, you are welcome to look at the original data here > > http://www.rcsb.org/pdb/explore/explore.do?structureId=2GL5 > > The data turned out to be twinned ( I also have other datasets - all with certain degree of twinning) and now I am trying to re-refine and then re-submit the more correct structure to the PDB. > > > > Yury > > On Thu, 10 Feb 2011, Garib N Murshudov wrote: > >> Maximum theoretical drop R/Rfree for perfect twin from 30% is around 25% (i.e. it could go down to 5%). However it could only happen only if twinning is perfect and there is no pseudo rotation parallel to twin operator. >> Hypothetical case it can happen if you have refined one crystal structure at sufficiently high resolution till (almost convergence) and another crystal is twinned but otherwise perfectly isomorphous to the first crystal and you take coordinates from the first crystal and refine against the second crystal. >> >> regards >> Garib >> >> >> On 10 Feb 2011, at 20:14, Patskovsky Yury wrote: >> >>> Dear all, >>> >>> >>> Twin refinement has yielded Rwork/Rfree values of about 0.10/0.12 for a nice quality 1.8A dataset (Rmerge 6%, space group I4, twin fractions 0.6/04) and almost the same R/Rfree (0.095/0.115) for another 1.5A nice quality data set (Rmerge 6%, space group I4, twin fractions 0.74/0.26). Refinement of untwinned data resulted in Rfree of ~32% and ~22% respectively. REFMAC and PHENIX both have produced the same results and almost identical R factors, which are suspiciously VERY LOW for this resolution of data. Twin refinement in REFMAC has produced exceptional quality maps even for 1.8A data (they look rather like 1.2A maps) - I can not tell the same for PHENIX - maps were looking worse (may be someone has a better idea why). >>> Normally twin refinement results in lowering R-factors - say, the drop in R from 30% (without twin refinement) to 20% (with twin refinement) would be considered normal, however we can see the drop from 32% to 12%. >>> I wonder if anyone else has experienced similar problems and what would be the most reasonable explanation for that. >>> >>> >>> Thank you >>> >>> Yury >>> >>> >>> >>> >>> >>> Yury Patskovsky, Ph.D. >>> Associate, >>> Dept of Biochemistry >>> Albert Einstein College of Medicine >>> 1300 Morris Park Ave >>> Bronx, NY 10461 >>> phone 718-430-2745 >>> [log in to unmask] >> >>