[phenixbb] R-factor problem !!!

Pavel Afonine pafonine at lbl.gov
Sun Jul 29 20:15:46 PDT 2007

Hi John,

john kryst wrote:
> Phenix.refine gave a R and Rfree of  0.1929 &  0.2512 ( rmsd bonds = 
> 0.006 )  respectively. When i format in CNS it gave R and Rfree 0.2282 
> & 0.2794 ( rmsd bonds= 0.010865 ). Am i doing any mistake ??

The difference in R-factors is because of the following (at least):
1) Different bulk-solvent modeling (targets, protocols) (currently we 
have the most sophisticated algorithm);
2) Different anisotropic scaling protocol (tightly integrated with the 
bulk-solvent modeling);
3) Different mask calculation parameters;
4) Different way of TLS handing (if used at all in your case);
5) Different form-factors values;
6) If you use H atoms, other programs may have their specific ways of 
treating them at input;
.... I can make an endless list of why the values can be different. 
Certainly, the bulk-solvent and anisotropic scaling are two the most 
probable candidates for making this difference.

In general, I would NEVER use another program to generate statistics 
file. Even not because different programs produces slightly different 
results (which is normal), but because there are myriads of subtle 
parameters that needs to be taken care of and adjusted to PROPERLY 
reproduce a result when transferring from one program to another. 
Otherwise you are bound to end up with the difference in numbers like 
you got.

As a temporary work around I would just get a template PDB file header 
and copy in necessary numbers from the output phenix.refine file.

As for the different stereo-chemistry statistics values (bonds, angles): 
we recently discovered that we were printing out different values from 
what the users normally expect to see (I wouldn't call it a bug, 
though). Previously we we printing mean errors. We corrected it and now 
we print rmsd values, as people expect. Please get the latest PHENIX to 
have this fix.


More information about the phenixbb mailing list