hmmm - it is quite difficult to do a good analysis of your problem,
remotely. You've tried the enantiomorphic space group I4(3), just to be
sure? In principle, the molecular replacement solution given by Phaser
sounds good, but this is no proof of whether it's correct. What sounds
good is, that you have high Z-scores, the packing looks good to you and
the starting R-factor is quite low, not high! For a first model, the
working-R and Free-R are usually very close (the initial unrefined model
explains the whole set and a random subset equally well), and it is
normal, that the Free-R rises in the first xyzB refinement cycles (the
free set "decouples" from the working set, so to say). You don't say how
much it rises, but make sure that you use _very_ tight geometry
restraints at this resolution to reduce overfitting. I suggest a
starting weight of 0.01 or even 0.005 for the X-ray term, resulting in
final RMSD values for the bonds of ~0.010-0.012 (there is no exact
rule). Your could also use tight NCS restraints for the two molecules to
even further reduce any potential overfitting. In contrast, you can
leave the B-factor restraints as they are, or even loosen their
restraints (I use always values of 2.0, 3.0, 4.5, 6.0 - you can find
them in REFMAC under the Geometrical Parameters tab)! On one hand, this
gives the model more freedom which could potentially result in more
overfitting, on the other hand, however, this additional freedom, at
least in my experience, effectively _reduces_ model bias by allowing
wrongly placed atoms to vanish. In general, for molecular replacement,
never believe any "solution" until you see the resulting electron
density maps. For a true solution, the electron densities should tell
you, which parts are wrong and which are missing. At 2.9 A, resolution,
this might be difficult. I would suggest that you look at the
NCS-averaged electron density map in Coot.
However, there could be a lot of other problems ...
I would suggest that you could find a local crystallographer at Berkeley
that helps you on site.
I hope that helps a little bit.
Scott Coyle wrote:
> I'm an undergraduate and recently crystallized and obtained 2.9A
> diffraction data for a protein which is predicted to fold into a WD40
> 7-bladed beta-propeller structure (which has been crudely verified by
> cryo-EM by another lab). The space group appears to be I4(1) with unit
> cell parameters 118.936 118.936 85.456 90.000 90.000
> 90.000. Using a number of different search models (which I trimmed and
> aligned to my protein's sequence using Chainsaw) I have obtained a
> number of MR solutions placing 2 molecules in the AU with Phaser with
> high Z-scores (ranging from Z=9 to 12) that seem to pack together
> nicely, so I was hoping to use this technique to solve my structure.
> However, the initial Rfree for my best solution is relatively high
> (0.49) and all attempts to refine the structure result in the Rfree
> blowing up almost immediately. This makes me worry that the maps I'm
> generating may be too model-biased to use to generate a solution. I've
> tried using Prime and Switch to remove model bias but the resulting map
> looks worse than the starting map. As the predicted structure possesses
> so much radial symmetry (7-fold), I'm worried that my MR solutions will
> never be oriented correctly enough for me to be able to build a model.
> If anyone has any suggestions for tackling this kind of molecular
> replacement woe, I would greatly appreciate it. Otherwise I guess I'll
> just plan to collect experimental phasing information sometime in the
> near future.
> I'm not sure if this is the right place to be asking this question,
> perhaps you guys could direct me elsewhere.
Paul Scherrer Institut
Biomolecular Research, OFLC/110
CH-5232 Villigen PSI, Switzerland
E-mail: [log in to unmask]