On Wed, Apr 20, 2011 at 3:14 PM, Yuri
I just noticed that the latest model I am working on has only 2% of reflections marked as free R after refinement! I usually give the GUI the .sca files out of HKL2000 that do not have an Rfree array pre-defined. The GUI lets you know about the lack of it and asks me if I want generate a test set of 10%. I say yes. However the pdb header shows the only 2.14% are chosen for free R. I see that some of my other models (lower resolution) have test arrays of around 7-10% as I would expect.
This is because Phenix has a default maximum of 2000 reflections in the test set. This is considered enough for all of the necessary calculations to be robust, apparently (the maximum likelihood targets depend on the test set somehow, but won't work properly if the reflections are too sparse).
How can I explicitly define the size I want?
In phenix.refine, click the "Options..." button in the section labeled "X-ray data and experimental phases", look for a field labeled "Maximum number of reflections in test set", and delete the value. You can also do this from the reflection file editor, via the "R-free flags generation" button. -Nat