Hi Pavel, I agree that you bias R-free after the real-space refinement and a quantification would be very interesting. My point was that we normally do not calculate R-free after real-space refinement, but after reciprocal space refinement. Here the bias is removed again. So, the problem with bias seems a matter of principle.
In practical terms we have to choose the best option: 1) Refine against maps with missing reflections and the possibility of artefacts that lead to suboptimal results. This keeps R-free unbiased, even directly after real-space refinement. 2) Refine against maps with test reflections set to Fcalc. This biases against the current model, but should have less artefacts. This too keeps R-free unbiased directly after real-space refinement. 3) Refine against maps with all reflections. This should give the best fitting results, but does introduce bias to the test set. However, this bias is lost after (converged!) reciprocal-space refinement. I think that for the final R-free these are equivalent. Cheers, Robbie > -----Original Message----- > From: Pavel Afonine [mailto:pafon...@gmail.com] > Sent: Thursday, April 24, 2014 17:27 > To: Robbie Joosten > Cc: CCP4BB@JISCMAIL.AC.UK > Subject: Re: [ccp4bb] coot problems to decrease R FREE > > Hi Robbie, > > > I agree with Ed that you probably should not leave out your free > reflections in real-space under normal circumstances. > > > Phoebe is right: you should not use free-r reflections in calculation of map > that you are going to use for real-space refinement. Otherwise you will be > biasing Rfree: the strength of the effect depends on how much model you > will be real-space refining. I can clearly demonstrate it with a simple cctbx > based script that I can post later of anyone interested. Ed's point about > zeros > is a good one, but here we are talking about specific set of non-zero > reflections (free reflections) that we do not want to "see" any model, any > refinement. If by "under normal circumstances" you mean real-space > refinement of only a tiny fraction of the model (say a residue or a few > residues) followed by reciprocal space refinement then this may not visibly > impact the Rfree; however, still, technically this will make free-r > reflections > "see" your model and some refinement, which isn't cool I think. > > All the best, > Pavel
