Dear Bernhard,
I think, the main difference of an unmodelled part between the mask bulk
solvent correction and the Babinet bulk solvent correction is, that the
mask approach can use quite detailed structural information, whereas the
Babinet approach uses only two additional scale factors. Let's
illustrate this for a model with an unmodelled part:
In the mask bulk solvent correction, the solvent mask fills the region
outside the modelled part and thus lowers the local contrast of the
unmodelled part. For the unmodelled part, it would be better if there
wouldn't be any bulk solvent density (BUSTER allows this). Then, the
superior scaling of the mask bulk solvent approach without reducing the
local contrast should really help with the interpretation of the
unmodelled part.
However, masking out the unmodelled part shouldn't be too detailed,
since the hole left in the bulk solvent mask will return as positive
difference density and can be heavily biased towards the atoms that were
used to exclude the solvent mask. I've seen this in the very old X-Plor,
where setting the occupany to "0" of a part of the model took it out
from the model's Fcalc but not from the solvent mask calculation,
leading to positive difference density exactly around the "omitted"
atoms + mask radius.
Something like this could not happen with the Babinet approach, since it
only affects overall scale factors. But here, the effect is quite
counter-intuitive!
If the unmodelled part is well ordered (but just missing), it's
contribution will be missing from the model's Fcalc at all resolutions
and thus will affect the overall scale factor of the model Fcalc, only,
but not the additional Babinet scale factors. If the missing atoms are
not accounted for in the overall scaling, this could decrease, or
underestimate, the signal of the difference density everywhere, which is
a general problem, but not specifically related to the Babinet solvent
correction.
If, however, the unmodelled part is less well ordered (which is the more
common case), it's contribution will mainly affect the model's Fcalc at
low resolution. If the overall scaling is dominated by the high
resolution terms (which is usually the case), the unmodelled part will
not have an effect on the overall scale factor of the model's Fcalc,
leaving only the Fcalc at low resolution somewhat too low. Since
lowering the model's Fcalc scale factor at low resolution is the main
purpose of the Babinet solvent correction, this will lead to an
_underestimation_ (!) of the Babinet bulk solvent contribution scale
factor (ksol). This, in principle, should increase, or overestimate, the
signal of the difference densities at low resolution, which might also
help with interpretation of the unmodelled part.
But all these scaling effects should be small, unless a substantial part
of the model is missing.
Best regards,
Dirk.
On 12.01.2015 11:09, Bernhard Rupp wrote:
What still evades me is, why exactly is the Babinet immune to these
effects of excluding/masking-out unmodelled parts?
The Babinet correction is also a function of the MODELLED part, just
the opposite sign. So an incomplete model
de facto equals an over-estimated solvent. Is it just the high
effective dampening of this correction (B ~200) as
Pavel said that makes it less susceptible because the higher
resolution reflections are less affected and
therefore have a chance to correct/overcome the inadequate (implicit)
masking?
Best, BR
*From:*CCP4 bulletin board [mailto:CCP4BB@JISCMAIL.AC.UK] *On Behalf
Of *Eleanor Dodson
*Sent:* Sonntag, 11. Januar 2015 17:05
*To:* CCP4BB@JISCMAIL.AC.UK
*Subject:* Re: [ccp4bb] Bulk solvent
Yes. If the model is incomplete it is obviously not sensible to use
the mask based solvent - you will tend to lose the unmodelled
features. It also gives unrealistically low R factors for a crystal
with high solvent content. However the best test would be to analyse
maps and try to decide if & when the different procedures work best..
A project for a student dissertation perhaps??
Eleanor
On 9 January 2015 at 20:13, Roberts, Sue A - (suer)
<s...@email.arizona.edu <mailto:s...@email.arizona.edu>> wrote:
I always try both methods - usually there is little difference. However,
For CueO (multicopper oxidase) where there are about 25 disordered
(unseen) residues in a loop, using Babinet scaling instead of the
default refmac scaling reduced the R factors by about 2% (both R and
Rfree) and improved the quality of the maps substantially.
From Dirk's comment, I'd guess this is because the mask-based solvent
model is putting solvent where there is (disordered) protein, which is
different from real bulk solvent.
Sue
Dr. Sue A. Roberts
Dept. of Chemistry and Biochemistry
University of Arizona
1306 E. University Blvd, Tucson, AZ 85721
Phone: 520 621 4168 <tel:520%20621%204168>
s...@email.arizona.edu <mailto:s...@email.arizona.edu>
-----Original Message-----
From: CCP4 bulletin board [mailto:CCP4BB@JISCMAIL.AC.UK
<mailto:CCP4BB@JISCMAIL.AC.UK>] On Behalf Of Armando Albert
Sent: Friday, January 09, 2015 12:56 AM
To: CCP4BB@JISCMAIL.AC.UK <mailto:CCP4BB@JISCMAIL.AC.UK>
Subject: [ccp4bb] Bulk solvent
Dear all,
Is there any reason for using Babinet scaling for bulk solvent
correction instead of mask based scaling?
Armando
--
*******************************************************
Dirk Kostrewa
Gene Center Munich, A5.07
Department of Biochemistry
Ludwig-Maximilians-Universität München
Feodor-Lynen-Str. 25
D-81377 Munich
Germany
Phone: +49-89-2180-76845
Fax: +49-89-2180-76999
E-mail: kostr...@genzentrum.lmu.de
WWW: www.genzentrum.lmu.de
*******************************************************
