Hi Kerstin, "Should I refine a low-resolution dataset 6 A": usually we refine a model of the crystal structure against measured data set.
Besides, technically you can do this, and I hope other replies will help you set correct expectations about the realities of the process. At this resolution you are not expected to have any nonsensical geometry outliers (such as rotamer, Ramachandran plot, cbeta, etc) as you will not be able to justify them against the map (map simply does not contain details necessary for this). You can obtain outlier-free model in Phenix that will also fit the map (and that will include refinement of individual coordinates). I'm sure you can do it with other tools available to you. Re similar resolution cryo-EM maps and corresponding models.. Well, in X-ray case maps are always model biased because model phases (or their combination with experimental phase information) are always used to calculate the maps. In cryo-EM this is not the case: map is what you refine against and obviously no model phase information is used to calculate this map. This is why the X-ray case is likely to be more tricky provided the resolution is the same. Pavel On Tue, Jun 9, 2015 at 2:19 PM, Kerstin Keller < kerstin.keller.1...@outlook.com> wrote: > Hi Experts, > > Last year I collected a dataset at 6 A of 2600 aa multi-domain protein. > > Resolution range 50 - 6 (6.7-6) > Wilson B 370 A^2 > Reflection 34120 (12240) > Rmeas 0.2 (4.4) > Rpim 0.06(1.2) > I-sigI 9 (1) > CC1/2 0.9 (0.5) > > I have the following questions in my mind. > > 1. Does it make any sense to solve the structure at _this_ resolution? It > is not completely novel protein, there are known structures with about 54 % > identity to this. The fold is known to be the same. > > 2. Doing a molecular replacement with "Phaser" using EM-model gives me a > unique solution. And I can see a reasonable electron density map. Tried > again with AMORE - The amore-build-output model also gives the same > solution. In both cases, the solutions are unique. There are no so-called > translational symmetry. > > 3. If I do a Refmac restrained refinement, though I get R/Rfree in around > 30.1/35.5 the stereochemistry is very poor (18 % outliers). I had to enable > tight WEIGHT MATRIX (1e-7). Here at this resolution does it make any sense > to a restrained refinement? > > 4. If I do only a rigid body with Refmac, the R-factor/Rfree are at around > 41 %, and in many places model does not fit density. When I manually > correct these and refine there is basically no change in R-factor/R-free > (it even worsens in cases). > > >From: Bernhard Rupp (Hofkristallrat a.D.) <hofkristall...@gmail.com>>Date: > 27 April 2015 at 21:54 > >Subject: Re: [ccp4bb] 3BDN, 16.5% Ramachandran Outliers!!!!! > >To: CCP4BB@jiscmail.ac.uk > >What we cannot tell sans supporting density is whether it is a more > accurate model, although I have>rarely seen an improvement in geometry > giving worse density fit. Usually a mess remains a mess - > >there is (at this resolution) no free lunch. The key question is again – > does the model justify > > 5. I did try DEN, reference model restraints, PROSMART(refmac) etc which > have no improvement at all - R/Rfree stuck at 41 %. > > 6. Since my group is EM-group, I wonder when EM-maps of 6 A are published, > why are X-ray data at the same resolution not being published? What happens > to these datasets? > > 7. Can I _just_ do a molecular replacement and just mutate residues (based > on a sequence alignment - There are large numbers of deletions and hence > sequence registers are different/unknown) and deposit it as a model in the > PDB? Should I put the side-chains or it is meaning less at this resolution? > Why in the EM-field they are allowed to deposit such coordinates with > side-chains? > > 8. As hofkristall...@gmail.com points out > > >Particularly in Molecular Replacement structures, and here particularly > in those with multi->segment/domain models, there are almost always parts > that fit well and others>that fit poorly - with simply not enough data at > the given resolution to improve the poor parts>sans additional phase > information. Bias issues have been discussed and need not be iterated here. > Since my protein also has multiple catalytic domains, some of them better > resolved and others terrible. What about bias here at 6 A resolution? Very > large problem, indeed. > > Apologies for the long email, and any suggestion will be gratefully > received. > > Many thanks, > Kerstin > > > > ---------- Forwarded message ---------- > From: Bernhard Rupp (Hofkristallrat a.D.) <hofkristall...@gmail.com> > Date: 27 April 2015 at 21:54 > Subject: Re: [ccp4bb] 3BDN, 16.5% Ramachandran Outliers!!!!! > To: CCP4BB@jiscmail.ac.uk > > I’d be very careful at judging low resolution structures. This is a tricky > businessrequiring a lot more info than just the PDB validation report. The > 3+ to 4 Aresolution range is a particularly deceptive one: The > crystallographer does not have much data given the model parameters > (perhaps consulting his figure showingdeterminacy for coordinate refinement > might help) > http://www.ruppweb.org/Garland/gallery/Ch12/pages/Biomolecular_Crystallography_Fig_12-11.htm > At > this resolution one has about enough data to keep enthusiasm up but at the > same time it isnot quite yet bad enough to throw up the hands and admit > that that one is de facto modelling with a few X-ray restraints (i.e. > data), requiring correspondingly suitable refinement protocols (and > discipline,aka mental restraints in addition to stereochemical > restraints). One is easily spoiled by looking exceptional 2A structures of > huge complexes, butnature (I do not mean the journal but the same time > would not exclude it) is often cruel. Particularly in Molecular Replacement > structures, and here particularly in those with multi-segment/domain > models, there are almost always parts that fit well and othersthat fit > poorly - with simply not enough data at the given resolution to improve > the poor parts sans additional phase information. Bias issues have been > discussed and need not be iterated here. Pavel is correct in pointing out > that a model with better geometry is also a more plausible model.What we > cannot tell sans supporting density is whether it is a more accurate model, > although I haverarely seen an improvement in geometry giving worse density > fit. Usually a mess remains a mess - there is (at this resolution) no free > lunch. The key question is again – does the model justify the specific > conclusions drawn from it? If a poor model is better than no model at all, > be it, as longas this is recognized and not used as an excuse for careless > work. Facile dictu, difficile factu. >
