To add to Tim's references, the algorithms that pdb-redo uses to deal with Rfree and possible bias are described here: https://journals.iucr.org/d/issues/2012/04/00/ba5174/index.htmlIf you happen to have a very small dataset (< 5000 reflections) pdb-redo does 10-fold cross validation and returns
Dear Qixu Cai,
it is a misconception that Rfree was not free anymore. The gap does
not close when you swap the reflections in Rfree set. The 'free'
reflections are freed when you do proper refinement, i.e., when the
value of the target function becomes stationary.
See
Dear Robbie and Tim，
Thanks a lot for your reply. I just do not understand that if we can change
the R free set during different rounds of refinement, the gap between Rwork
and Rfree would be small, and Rfree would be meaningless, as R-free set is
not "free".
Best regards,
Qixu Cai
Email:
Hi Qixu,PDB-REDO tries to have a minimum number test set reflections to reduce the error margin in R-free. As Tim says this is not a problem but if you reach out privately we can change your calculation to use your current testset. That is not recommended though.Cheers,RobbieOn 28 Apr 2023 04:51,
Dear Qixu Cai,
the sets do not have to be identical, as long as refinement reaches a
steady state.
Best regards,
Tim
Am 28.04.2023 04:51, schrieb Qixu Cai:
Dear all,
I'm using the PDB-REDO server to refine my structure. I found that
PDB-REDO said that 5% R-free set is too small and it
