[buster-discuss] Exploding R-factors / adjustment of X-ray weight ?

[Krey.Thomas at mh-hannover.de]

Dear Clemens and Gerard,

Thanks a lot for your suggestions – yes, your suspicion about the incomplete model was right – I solved the issues by placing more molecules per AU (currently refining around R of 0.23), which did not drop out of the first MR runs.
I seem to remember that there was an option (or a macro?), where one could specify the number of missing atoms. Would that have helped in this case? In any case, is there a good rule of thumb, which percentage of the AU one needs to cover to avoid such a behaviour?

All the best
Thomas


Prof. Dr. Thomas Krey
Hannover Medical School  
Institute of Virology
Structural Virology Group
Carl-Neuberg-Str. 1          
D-30625 Hannover
phone: +49 (0) 511 - 532 4308
email: krey.thomas at mh-hannover.de
 

On 2/13/17, 7:56 PM, "Clemens Vonrhein" <vonrhein at globalphasing.com> wrote:

    Dear Thomas,
    
    On Sat, Feb 11, 2017 at 09:29:34AM +0000, Krey.Thomas at mh-hannover.de wrote:
    > I am refining a 2.6A structure with a relatively large hexagonal
    > unit cell (~260x260x200). There is no indication for problems in
    > indexing (e.g., twinning) nor for pseudotranslation and the MR
    > solution with 5 molecules per AU looks relatively clear (TFZ of
    > ~31).
    
    Is this the number of mol/asu you expect? How homologous/identical and
    complete is your search model (i.e. is it just main chain? or
    poly-ALA? or poly-SER?).
    
    > Nevertheless, the refinement explodes to R-values of about 0.85,
    
    This sounds very much like an incomplete (or very poor) starting
    model.
    
    > but surprisingly the corresponding maps look quite reasonable (so
    > the molecules do not seem to be placed simply wrong!).
    
    Do you see where you need to correct your current model or where you
    need to build new model? This would be a very good indication of the
    predictive power of your current phases.
    
    Can you send us the reciprocal space CC plot of that refinement,
    e.g. after
    
      graph_autobuster_recipCC -d your-output-dir
    
    This might enable us to see what the reason for those high R-values
    is.
    
    > My question is, whether these high R-factors can be temporary and
    > related to the adjustment of the X-ray weight and should go down at
    > some point or if there is something more basic that I am doing
    > wrong?
    
    They can be temporary, yes. However they are unlikely to be related to
    anything to do with the X-ray weight.  We've had multiple cases with
    poor R-values (and increasing with refinement) while the reciprocal-
    space CC plots showed values that stayed below a ceiling that was very
    much less than 1.0, e.g typically 0.5-0.6 instead of the 0.8+ values
    normally seen with a good model. Their meaning was therefore that the
    model was systematically incomplete - that is why we want to take a
    look at your RecSCC plots, to check whether this is the case with your
    model.
    
    Best wishes
    
    Clemens & Gerard (for buster-develop)