[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [ccp4bb]: low res dataset, need help with NCS constraints

***  For details on how to be removed from this list visit the  ***
***    CCP4 home page http://www.dl.ac.uk/CCP/CCP4/main.html    ***

Susan Heffron wrote:
> ***  For details on how to be removed from this list visit the  ***
> ***    CCP4 home page http://www.dl.ac.uk/CCP/CCP4/main.html    ***
> Hi all!
> I could use some advice about refining a model with some fairly low-
> resolution data.  Here's my situation:
> I have a relatively low resolution dataset (~3.2A) in P212121, which
> consists of about 12,000 reflections.  My model is a molecular
> replacement solution, with two molecules in the a.u. and a total of
> about 6500 atoms.  I have been having trouble getting the refinement
> (using CNS) to proceed.  My R-free is still above 30%.  One concern I
> have is that the data-to-parameters ratio is less than 1 (12,000
> reflections, compared to 19,500 x-y-z positions).  I have been using NCS
> restraints, but wonder if constraints would be better, in order to
> improve the data-to-parameter ratio.  Am I correct in thinking about it
> this way?
> If I do use NCS constraints, I have the complication that my protein is
> a multi-domain molecule that has flexibility in the orientations of the
> domains to each other. So, it would have to be constrained
> domain-by-domain.  Which programs can handle this?

 This isnt a problem with PROTIN/REFMAC

 You assign NCS as residue spans and each one will have a slightly
different rotation matrix 

NONX  2 CHNID A B NSPANS 1  100 150 2
etc etc

Then REFMAC will automatically do what you require - NCS for each domain
will be done independently.

> One other possibility I've considered is to restrain the structure to a
> higher resolution copy with harmonic restraints.  This would require
> superimposing a higher-res structure, again domain-by-domain, and then
> refitting any loops that are obviously different in the current maps.

 Doubtless a good idea - but as far as I know not really implemented in
a program. There are serious theoretical problems in weighting the
different sources of information! 
After refining int 3.2A structure you can probably improve the final
model by checking which parts have remained identical to the high resln
model and maybe even fixing those regions to the hr model and excluding
them from the refinement;  and simply refine the changed regions.. But
it all gets rather messy, and if you are solving this structure to see a
ligand not really very important.. Achieving a Free R under 30% does not
mean you have 100% reliable coordinates with such limited data!

 Eleanor Dodson