Hi Liuquing,
> For example,first,the resolution is 2.25,R-free is 30,completeness is 90%.
> Then I changed resolution to 2.35,R-free became 28.0,completeness is 91%.
> When I changed resolution to 2.45,R-free became 27.5,completeness is 92%.
> So,when resolution from 2.25 to 2.35,R-free goes down a lot,but
> resolution from 2.35 to 2.45,R-free nearly the same,is that meaning
> resolution should around 2.35?
This is pretty expected. Just plot R-factor vs resolution for some good
amount of structures from PDB.
> Should I change resolution during refinement?
Only if you have a very *good* reason for doing so.
> And is there any good software for refinement?
> <http://spaces.live.com/?page=HP>
Here are few hints of what's out there:
1) SHELXL (great for refinement at high resolution, comprehensive H
atoms handling, twinning; slow since does not use FFT, poor bulk solvent
model, no ML targtes);
2) CNS (uses simulated annealing = large convergence radius; good bulk
solvent model, twinning, uses various ML targets; high-resolution
refinement is not supported, no TLS);
3) REFMAC (TLS, high-resolution refinement, uses ML targets; bulk
solvent modeling often unstable);
4) PHENIX (refinement at any resolution is supported: from ultra-high (<
0.9A) to low with sophisticated rigid body refinement (very big
convergence radius); robust TLS (never gets "unstable"), robust bulk
solvent model (always finds best parameters), simulated annealing, twin
refinement, H handing, alternative conformations; torsion angles
dynamics is not yet implemented).
To find out more about each option -- do Google search.
Cheers,
Pavel.
|