[phenixbb] Bias in Rfree
pafonine at lbl.gov
Tue Jul 2 08:07:35 PDT 2013
phenix.refine has an option to generate free-r flags in thin shells with
20 being the default number of shells (it is a user changeable
parameter). It's not exactly what you quoted below but my understanding
it is meant to address this potential bias.
On 7/2/13 1:32 AM, "Weiergräber, Oliver H." wrote:
> while refining a structure with NCS, I was re-considering this paper from Michael Chapman's group:
> Fabiola F, Korostelev A, Chapman MS. Bias in cross-validated free R factors:
> mitigation of the effects of non-crystallographic symmetry. Acta Crystallogr D
> Biol Crystallogr. 2006 Mar;62(Pt 3):227-38.
> In essence, they suggest to exclude from the working set reflections that are
> (1) related to a test reflection by NCS
> (2) adjacent to a test reflection and its NCS mates in reciprocal space
> They have implemented this strategy in a CNS module (R-free2005), but the software is no longer actively supported.
> Does anybody know if something similar has ever been implemented in other software packages such as phenix?
> Or am I missing any significant downsides of the approach - apart from a reduction of the working set?
> Best regards,
More information about the phenixbb