Does phenix.refine take NCS into account when picking the Rfree set? Thin shells, or other strategy...
The short answer is no. We had long internal discussions about this without reaching a final conclusion. In the long run we may integrate alternative R-free set distributions in phenix.refine; in the meantime you have to use an external program (e.g. dataman or sftools). BTW 1: just to prove that we have thought about this subject, we have a simple tool for "visualizing" distributions of R-free flags: iotbx.r_free_flags_accumulation your.mtz It writes a file with x,y pairs which you can plot with e.g. Excel or any other 2D plotting program. BTW 2: The free reflections aren't as free as one may think. When using a maximum likelihood target (any program), the free reflections are used in the estimatation of the error model. This estimation becomes more tricky if the reflections are distributed in thin shells. I.e. if you go to thin shells you may in fact account to some degree for the NCS, but at the same time you compromise the error model estimation. Ralf