On Wednesday 30 June 2010 11:57:45 am Pavel Afonine wrote:
> Hi Ethan,
>
> On 6/30/10 11:43 AM, Ethan Merritt wrote:
> > On Wednesday 30 June 2010 11:15:51 am Pavel Afonine wrote:
> >
> >> Just a remark/precaution: if you remove ANISOU records then you will
> >> basically invalidate the refinement results (the refined model).
> >>
> >
> > If that is true then I would say there is a deeper problem.
> >
> > The description of the TLS groups in the header records should be
> > sufficient to regenerate the ANISOU records when needed. If this
> > is not already the case, then I suggest that we, as a community,
> > need to work out how to make it the case and then lobby for that
> > solution to be mandated by the PDB.
> >
>
> Specifically to phenix.refine: phenix.refine outputs total ADP into
> ANISOU records and its isotropic equivalent into ATOM records. If you
> just remove ANISOU records you will still have the total B-factors in
> ATOM records. Now, what happens if you restore back the TLS contribution
> using TLS matrices in PDB file header is that you will get ANISOUs
> containing only TLS contribution only (I presume), and B-factors in ATOM
> records will still be the full ones. So, total mess, obviously.
You can call it a total mess, and I even agree with you, but it is
consistent with what the PDB currently mandates as I understand it.
However if you prefer to further modify the regenerated ANISOU records so
that they also contain the contribution from a residual Biso term
that is a trivial additional step. The residual Biso is the difference
between the value for B in the original ATOM record and the value derived
from the regenerated ANISOU records.
> Anyway, post-refinement manipulation on a PDB file is bad.
Heh. Tell that to the PDB.
cheers,
Ethan
> > But I think it's not that bad. I have had good results from reproducing
> > the reported R factors from the TLS header records plus the ATOM records.
> > Admittedly, the PDB files I used for testing were refmac refinements
> > rather than phenix refinements.
> >
>
> This is true. Because most likely the PDB files you tried were
> containing residual B-factors in ATOM records and the TLS contribution
> stored in REMARK 3.
> I was getting the same results too:
>
> Afonine et al., J. Appl. Cryst. (2010). 43
> phenix.model_vs_data: a high-level tool for the calculation of
> crystallographic model and data statistics.
>
> >> Make
> >> sure you recompute the R-factors after removing ANISOU and be prepared
> >> to see (much) higher values.
> >>
> >
> > Well, that depends on what program[s] you use to recalculate the R values.
> > If you actually use the TLS information in the header records then it
> > should come out close to the original values.
>
> See reference above.
>
> Pavel.
>
>
--
Ethan A Merritt
Biomolecular Structure Center, K-428 Health Sciences Bldg
University of Washington, Seattle 98195-7742
