On Thu, Apr 8, 2010 at 3:26 PM, Daniel Bonsor <bon...@bbri.org> wrote:
> Following my previous question, there was something wrong with the staring > model for molecular replacement. Now that is sorted, I have 8 complexes in > the ASU. After a few rounds of refinement with NCS and isotropic Bfactors, > both the Rfactor and Rfree get stuck at 30% and 36%, respectively. > > I have only just noticed that I am trying to model ~43000 atoms with ~95000 > unique reflections (98% complete, at 2.58 Angstrom resolution). Is this the > reason why the R factors are stuck and I should start the refinement again > from scratch using overall Bfactors with NCS and switch to TLS once my > Rfactor is less than 30% and possibly reduce the number of reflections used > for Rfree (currently using the standard CCP4 5%) > I don't think it's a parameters-to-data-ratio problem - the combination of NCS, geometry, and ADP restraints will help prevent overfitting, and the difference between R and R-free isn't unreasonable (they're just both too high). At 2.58A isotropic B-factors are very appropriate, and switching to overall (per-residue?) will probably make everything worse. There is a long list of other things to check first, starting with twinning and/or other undiagnosed crystal pathologies such as pseudosymmetry. I'm not sure what to suggest about TLS refinement, but now would be a good time to try it and see if it helps. You could also try adjusting the strength of NCS restraints, but I doubt this will make a huge difference either way. Using a smaller test set will almost certainly not help anything. I've heard arguments that you can get away with less than 5% of reflections in the test set, if the resulting number is at least 2000, but personally, I'd keep 5%, if only to avoid the inevitable sniping by reviewers. -Nat
