On Sep 1, 2009, at 8:18 AM, Sam Stampfer wrote:
My PI (who is on vacation until next week) said that it was a problem if R-work was more than 0.06 lower than R-free. Any recommendations for me? Is it possible to reduce this difference? Is it worth it to rebuild the refined pdb if the R-work is this different from the R-free?
My refinement script is:
phenix.refine ss46_p3_cv.mtz ss46_rebuilt.pdb \ output.prefix=results/ss46_rebuilt_sa \ strategy=individual_sites+group_adp cif_link.params nag-nag.cif \ twin_law="h,-h-k,-l" \ simulated_annealing=true
Random ideas: - TLS (this sometimes has a huge effect on Rfree-Rwork) - Individual ADPs (may or may not help, but appropriate to try at that resolution) - What are your rms(bonds) and rms(angles)? If these are very high (or very low?) this may be the problem, and you need to optimize the weighting. - You should look at the other validation criteria as well (Ramachandran, rotamers, clashscore, etc.). -Nat ------------------- Nathaniel Echols Lawrence Berkeley Lab 510-486-5136 [email protected]