[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[ccp4bb]: Re: unmeasured data in map calculations

----- Original Message -----
From: "Jan Abendroth" <Jan.Abendroth@uni-koeln.de>
To: "ccp4" <ccp4bb@dl.ac.uk>
Sent: Wednesday, December 20, 2000 7:21 PM
Subject: [ccp4bb]: unmeasured data in map-calculations

> ***  For details on how to be removed from this list visit the  ***
> ***    CCP4 home page http://www.dl.ac.uk/CCP/CCP4/main.html    ***
> Dear all,
> with many apologizes (because it's mostly a a cns-related question) I
> like to submit a question to the ccp4bb. But it's soon x-mass and you
> might forgive this.
> My data set is quite incomplete in the highest resolution shell. When I
> calculate maps with the cns script model_map the sigmaA-weighted
> fofc-map (mFoDFc) has many "unreasonable" peaks at 3sigma-level. When I
> make cns to take Fc for the unmeasured Fo's the sigmaA-weighted fofc-map
> shows much fewer and much more reasonable peaks, simply this map becomes
> significanly better.
> This seems to make sense, but what is the physical/mathematical reason
> for this?
> How treats cns unmeasured hkl's in the "normal" mode in difference maps?
> Is there a related ccp4-option in refmac/fft?
> Thanks for any reply, kind regards and merry christmas
> Jan

Hi Jan,

the idea behind this is, as far as I have understood it, that the maximum
likelihood formalism should be able to produce reasonable estimates for the
"real" data (not necessarily the observed data). These (current) estimates
are model structure factors weighted by a factor for incorrectness of the
(partial) model, that is D*Fc. Unfortunately, only the amplitudes are
downweighted by this procedure, whereas the phases come directly from the
model and are still somewhat biased towards the model, depending on how far
the incomplete/incorrect model is kept apart from the observed data. By
substituting the missing Fo by D*Fc you have two opposite effects on the
resulting 2mFo-DFc electron density map (the mFo-DFc is not affected): the
positive effect is a much cleaner map, because any missing data in
reciprocal space will generally lower the quality of the map in real space
(especially when the missing data are not randomly distributed resulting in
a certain "shape" for the observed data, which will give a folding of your
real electron density map with the Fourier transform of that shape), the
negative effect is a somewhat increased model bias in your map. Some time
ago, I posted a related question on the CCP4BB, the feedback showed me, that
most crystallographers are not sure which effect is the more important one.
My suggestion would be (and this is what I will also do next time) to omit a
few well defined and a few poorly defined parts of your model and inspect
the electron density maps with and without Fo substitution for these "known"
parts. This procedure was suggested quite a long time ago by Rice to give
the user a feeling about model bias in maps calculated with combined phases
including model phases.

I wish all members of the CCP4BB a Merry Christmas and a Happy New Year!


Dirk Kostrewa
Paul Scherrer Institut      e-mail: dirk.kostrewa@psi.ch
Life Sciences                 phone: +41-56-310-4722
OSRA/007                     fax: +41-56-310-4556
CH-5232 Villigen PSI      WWW: www.sb.psi.ch