Hi

I highly recommend reading the two Brunger papers below: they will explain the important factors to take note of when refining low-resolution structures.  I found them very useful when I was learning about low-resolution refinement.

I would suggest using the deformable elastic network with simulated annealing in CNS, and then follow that with refinement in PHENIX using tight geometery restraints and doing TLS and grouped-B factor refinement.  Your problem is that at low resolutions, the data to parameters ratio is low, and you need to use a number of restraints along with good bulk-solvent and anisotropic corections in order to keep your R-factor and R-free values within 3-4%.  Perhaps you could send your PHENIX input file and we could see what you did and maybe point what needs to be corrected.

Hope that was helpful.  good luck!

Schroder, Levitt, Brunger, Super-resolution biomolecular crystallography with low-resolution data Nature. 2010 Apr 22;464(7292):1218-22

Brunger, AT et al.  X-ray structure determination at low resolution. Acta Cryst D. 2009 Feb;65(Pt 2):128-33



On Sat 07/09/11 4:59 AM , Qixu Cai <[log in to unmask]> wrote:

Dear all,

Recently, I refine two low resolution structures in refmac 5.5. Their
resolutions are 3A and 3.5A respectively.
For 3A structure, after MR by phaser and rigidbody refinement&restraint
refinement by refmac5.5, I got R factor 25% and R free 35%. And then
each time, after my model building in coot and restraint refinement by
refmac 5.5, the R factor stays 25%, but R free increases to 38%, even 39%.
For 3.5A structure, the R factor stays 27%, but R free increases from
37% to 42% after my slightly model building in coot.
Could you help me to find the reason?

Maybe the reason is the overfit of the structure? I found that new
version of refmac 5.6 has many new features for low resolution
refinement, such as jelly boy, secondary structure restraints. But I
don't know how to use these new features in old version ccp4i (6.1.13)?

I also used phenix.refine with the "reference model" ( I have high
resolution model for one domain of the low resolution protein) and
"secondary structure restraints", but it seams the same. Any suggestion?

BTW, is that simulator annealing not suitable for low resolution
structure? I used the simulator annealing method of CNS and
phenix.refine, but the geometry of the structure is always destroyed
seriously.

Could you help me?

Thank you very much!