Hi Rakesh,
providing everything else is good about your PDB file (for example, no
post-refinement manipulations done to the model), phenix.model_vs_data
should reproduce the R-factors, since it accounts for TLS (both, PHENIX
or REFMAC formatted), it does the advanced bulk-solvent and anisotropic
scaling (as Dirk mentioned is needed), plus, it will account for
twinning if available, handle multiple model PDB files, and structures
refined against neutron data and so on. (/J. Appl. Cryst./ (2010). *43*,
669-676).
~5% difference you mentioned is what you could expect if you ignore TLS
(and don't do proper bulk-solvent and anisotropic scaling - see
reference above).
To use:
phenix.model_vs_data model.pdb data.mtz
Let me know if you have questions.
Pavel.
On 7/6/10 8:28 AM, Dirk Kostrewa wrote:
Hello Rakesh,
yes, you are right. In contrast to all modern refinement programs,
sfcheck neither applies TLS nor the superior mask bulk solvent
correction (it applies the Babinet bulk solvent correction by
double-exponential scaling). As a result, the R-factors from sfcheck
are higher than those from the refinement program.
Annoyingly, I have to discuss/explain this discrepancy in R-factors
every time when I want to deposit a refined X-ray structure with the
Protein Data Bank ...
Best regards,
Dirk.
Am 06.07.10 17:05, schrieb Rakesh Joshi:
Hello,
I refined my structures using REFMAC5 ( using TLS in combination ).
When i used the SF-Check
program, my R-values are ~5% higher than in those reported by refmac.
Is this
due to the fact that SF-check does bulk solvent calculations
differently and
does not account for TLS refinement?
Also, when i used TLSANL to put in my total B factor in the pdb file
I only get
3 sets of numbers instead of( what i expected) 6 sets(ie. last three
entries have values="0").
For example:
ATOM 246 C6 DA C 13 -29.692 15.902 14.379 1.00
36.42 C
ANISOU 246 C6 DA C 13 3267 5459 5110 0 0
0 C
Cheers
Rakesh
Graduate student
Purdue University
