Hi Dirk,
thanks a lot for your feedback.
If you refine a macromolecular structure with phenix.refine using TLS
and isotropic B-factors, the resulting PDB file will have the
effective isotropic B-factors and their anisotropic corrections as
ANISOU cards from both the TLS components and the individual isotropic
B-factor component.
Just to make this sure once again. phenix.refine always writes out the
total B-factors and not split them. The total B-factor in phenix.refine
is defined as: Utotal = Utls + Ulocal + Ucryst. So, the ANISOU records
always contain Utotal and ATOM records contain isotropic equivalent of
Utotal which is trace(Utotal)./10000*8*pi**2/3 (correct me if I missed a
scale somewhere).
I think, this is very useful, because the PDB file has the
informations about the various B-factor components readily available,
both for non-crystallographers that are not aware of TLS and
for reproducing refinement results with other software.
This is what I had in mind. Plus, all the information about B-factors is
in ATOM and ANISOU records so: 1) you don't need any external programs
to put things together; 2) you can load your structure say into PyMol
and color it by B-factors, 3) you don't need to rely on PDB file header:
if you loose it you don't loose much of the information, etc...
Also, the latest development CCI Apps
(http://www.phenix-online.org/download/) has a new tool: phenix.tls that
allows you to go back and forth between both representations: total and
residual B-factors. It will be available in the next general PHENIX release.
However, if such a PDB file is fed into another refinement run with
phenix.refine, the program recognizes the ANISOU cards and
automatically switches to individual anisotropic B-factor refinement,
unless you explicitly state, that you wish to continue with isotropic
B-factor refinement by giving, for instance, a command line keyword
like adp.individual.isotropic=all or convert_to_isotropic=true (see
here <http://www.phenix-online.org/documentation/refinement.htm#anch18>).
Yes, it was in the past. Now it tries to be smarter about it. See below.
I think, this is not reasonable, since the vast majority of
macromolecular structures is refined at resolutions that do not
justify anisotropic B-factor refinement, and most users simply use the
default refinement parameters and are probably not aware of this
behaviour.
I agree. See below.
May I therefore suggest to the developers of phenix to revert this
behaviour, such that isotropic B-factor refinement is the default and
anisotropic B-factor refinement is only done, if the user explicitly
states that she/he wants it?
In the latest version (http://www.phenix-online.org/download/) the
default behavior of phenix.refine is (if no TLS is used):
- isotropic atoms in input PDB file are refined with isotropic ADP;
- anisotropic atoms in input PDB file are refined as individual
anisotropic if resolution is "high enough";
- anisotropic atoms in input PDB file are converted to isotropic and
then refined as individual isotropic if resolution is not "high enough".
The decision about whether the resolution is "high enough" or not is
made using this parameter:
switch_to_isotropic_high_res_limit=1.7
Of course a user can overwrite it by setting it to the value he/she likes.
If TLS is used: all atoms participating in TLS groups start having (or
keep having) anisotropic component.
I hope this clarifies things a little bit!
Pavel.
