On Sun, Jul 28, 2013 at 2:33 PM, Wei Shi
Thank you for the suggestions. I ran the refinement again with default settings + Rigid body+ Simulated annealing (Cartesian) + Optimize X-ray/stereochemistry weight. And the statistics improved! See below: R-work=0.2348, R-free=0.2855, Bonds=0.003, Angles=0.919. Ramachandran outliers: 2.5%, Ramachandran favored: 88.1%. Rotamer outliers: 10%, c-beta outliers: 0 clashscore: 9.06, overall score: 2.84
Now, I am wondering whether there is any suggestion about what to try and check next to further refine the structure. Thank you so much!
To start with, don't run rigid-body refinement or simulated annealing in future refinements. They're both very powerful tools for improving structures which fit the data poorly, but once the model is mostly correct they just take forever with little benefit (and in the case of annealing, sometimes make the structure worse). At this point I would go through the lists of outliers (i.e. residues with bad statistics) from validation and try to fix them individually. (You're probably not going to be able to correct every single problem - don't worry, it can be very difficult to get a "perfect" structure at this resolution.) Use Coot's Ramachandran restraints in real-space refinement, but be sure to inspect the conformation afterwards to see if it makes sense. Make sure you run the real-space refinement every time you adjust the model; this will make it much easier for phenix.refine the next time you run it. Without knowing more about your structure I don't know if you need to change your phenix.refine settings, but default settings plus weight optimization is a good start. *Do not* use the Ramachandran restraints in Phenix at this point. Also, as Pavel suggested, you may benefit from updating to the latest nightly build. -Nat