[phenixbb] test set count question

[Pavel Afonine]

Hi Filip,

> I am refining a structure with phenix.refine using a .mtz which 
> contains F/SigF and DF/SigDF (CCP format from xdsconv) and i have 
> something like 10000 unique reflections. The test set was created with 
> the uniqueify script. If I understand correctly in the manual it is 
> written that phenix.refine will separate the <F>/<SigF> into "F(+) 
> SIGF(+) F(-) SIGF(-)), instead of keeping the original non-anomalous 
> amplitudes". 

it's best to work with unmanipulated original data, which is Fobs(+) and 
Fobs(-).

  <Fobs> or "F/SigF and DF/SigDF" are derived from Fobs(+) and Fobs(-) 
with some information lost. That is you cannot restore the original 
Fobs(+) and Fobs(-) from <Fobs> or from "F/SigF and DF/SigDF". 
phenix.refine restores Fobs(+) and Fobs(-) from "F/SigF and DF/SigDF" 
for internal use making an arbitrary decision about the sign. 
phenix.refine cannot handle "F/SigF and DF/SigDF" and has no reason to 
do this.

Although I doubt it would make any visible effect on final refined 
structure.

> My question is about the "FREE R VALUE TEST SET COUNT" in the .pdb 
> after phenix.refine. If I understand correctly the number for the test 
> set will also be "doubled" 

Yes, it's a good idea to put "" around doubled since it's not precisely 
doubled because of +,- and singleton reflections.

> Logically both F+/F- after separation are kept in the test set is this 
> right? 

Yes, "+" and "-" reflections have the same test set value, if this is 
what you mean.

> And an impair number for the test set in this case means that only the 
> F+ OR F- was observed? Sorry if I ask a trivial question but I just 
> want to be sure I am using the program in the right way

"+" and "-" are individually measured reflections so I'm not sure why to 
hide this fact and count them as one reflection...

All the best,
Pavel