Hi Phoebe, I agree with Ed that you probably should not leave out your free reflections in real-space under normal circumstances. Especially the low resolution reflections will make your map look bad if set to 0.
Anyway, if real-space refinement is followed by reciprocal space refinement, and if this refinement is done to convergence, bias towards the test set will be 'forgotten' by the model. So any bias in real-space should go away. I guess that is one of the reasons why we always finish with reciprocal space refinement before depositing a model. Cheers, Robbie Sent from my Windows Phone ________________________________ Van: Edward A. Berry Verzonden: 24-4-2014 15:58 Aan: CCP4BB@JISCMAIL.AC.UK Onderwerp: Re: [ccp4bb] coot problems to decrease R FREE On 04/24/2014 08:33 AM, Phoebe A. Rice wrote: > If you are doing real-space refinement of the entire model against a map in > coot, isn't it also important to make sure that the Rfree reflections were > NOT used in calculating that map? Depending on how the map was made, that > info can be hidden in the fine print. > > ++++++++++++++++++++++++++++++++++++++++++ Not sure. Although this would seem to be the real-space analog of omitting the free reflections during reciprocal space refinement, there is actually a big difference. In reciprocal space, the free reflections can be simply ignored. In real space refinement against a map from which the Free reflections were omitted, you are still using the free reflections but with an incorrect value: zero. Then, if real space refinement really can bias a model to best-fit the reflections, all the free reflections will tend toward zero and your R-free for the model will go up. Then when you compare R-free after RS refinement with or without free reflections, you might incorrectly attribute the difference to bias when using them! eab > > Phoebe A. Rice > Dept. of Biochemistry & Molecular Biology > The University of Chicago > > 773 834 1723; pr...@uchicago.edu > http://bmb.bsd.uchicago.edu/Faculty_and_Research/ > > http://www.rsc.org/shop/books/2008/9780854042722.asp > > ________________________________________ > From: CCP4 bulletin board [CCP4BB@JISCMAIL.AC.UK] on behalf of Robbie Joosten > [robbie_joos...@hotmail.com] > Sent: Monday, April 21, 2014 4:40 AM > To: CCP4BB@JISCMAIL.AC.UK > Subject: Re: [ccp4bb] coot problems to decrease R FREE > > Dear Peter, > >> I'm a novice of coot and ccp4. Now I'm doing refinement using both refmac5 >> and coot.Here are some problems I'm facing. Really hope you can give me >> some suggestions. >> >> 1、THE RESOLUTION OF THE DATA IS 2.5 angstrom. After first refinement of >> refmac5 I got R factor which is 0.26 and R FREE which is 0.31. My question > is >> what the final R factor and R FREE should be after several rounds of >> refinement by refmac5 and coot. > As low as it can reasonably be. Which is of course a lousy target but the > purpose of model refinement is to make it as good as possible, not to > achieve a certain R-free. That said, at this resolution, the PDB average is > about 25% for R-free. > >> 2、At which map level（e/A3 or rmsd）should I refine the data by coot? > You refine against the whole map, but for viewing purposes you should change > the map contour level during building. For the majority of the map you > should be able to see stuff well above an rmsd of 1.0 for the 2mFo-DFc map. > At the difficult bits you may need to go below 1.0, but no so low that you > can fool yourself. The proper height for the difference map depends as a bit > on the situation. I usually just turn it down until obvious noise peaks > (small negative peaks in the solvent) appear. That typically ends up > somewhere between 3 and 4 rmsd. > >> 3、Can you give me some tips and strategies about how to use coot to >> decrease R free? now I just use some basic tricks such as fit density and >> Ramachandran plot to refine the data. > This is what I do for your kind of resolution: set the weight for refinement > in COOT to 50 or 40 (the default is 60) and switch on torsion restraints and > possibly Ramachandran restraints. Then I go through the entire structure > residue by residue (space bar) refining windows of three residues (with the > 't' key) and fixing everything that I can (that includes obvious solvent > molecules). Symmetry should always be switched on and NCS ghosts (if > available too). When that is done you can focus on the remaining difficult > bits by looking at the difference map peaks in COOT and at validation > reports from WHAT_CHECK, and MolProbity. > > So now for the bit where I plug my own stuff: you can try PDB_REDO > (http://xtal.nki.nl/PDB_REDO for the server, there is also a stand-alone > version) to take a lot of work out of your hands. It optimises your > refinement in Refmac and, rebuilds side-chain and tries to find peptide > flips that improve you Ramachandran plot and fit with the maps. You also get > a lot of validation information that may help you with further rebuilding of > your model. > > Cheers, > Robbie >> >> >> Best regards, >> >> >> Peter Chen >> >
