[phenixbb] anomalous difference map

Dale Tronrud detBB at daletronrud.com
Fri Oct 10 10:53:15 PDT 2014


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


On 10/10/2014 8:24 AM, Scott Horowitz wrote:
> So I have a followup question to this comment in the thread: what
> is the origin of this rule of thumb of 1 and 3 sigma that people
> quote for non-difference and difference maps that you can find all
> over the internet at various sites, but nobody references? I can't
> find any papers on it specifically, but it's quite possible I'm not
> digging far enough back in time. Can anyone give me a lead?
> 
> Thanks, Scott

   These particular contour levels are just "rules of thumb" as you
say.  The mistake of referring to them as "sigmas" gives the illusion
of some sort of statistical basis.  They are simply rmsd's and have no
mathematical justification for being used to assess the quality,
significance, or accuracy of a ligand identification (as Pavel noted).

   The 3 rmds contour level for Fo-Fc style maps is particularly
problematic.  The rmsd of the difference map for a model in the early
stages of refinement will be large while at the end it will be small.
 The same piece of density due to missing atoms will appear weak in
the early map and strong later.  Conversely the features visible in an
early map will correspond to striking errors in the model while
features in a final difference map that appear at exactly the same
strength will reflect tiny changes.

   This lack of calibration results in a continual hunt for poorer and
poorer water molecules because no matter how many you build there are
still plenty of "3 sigma" peaks in the difference map.

   The paper by Lang, Holton, Fraser, & Alber, recommended by Pavel,
discusses many of these issues.  They show that the real standard
deviation of electron density maps is much lower than these rmsd
values.  The problem isn't that density in a 2Fo-Fc style map less
than 1 rmsd isn't statistically significant but that it is hard to
distinguish between alternate models that all fit this density.
Telling the difference between a partially occupied ligand and
superimposed puddles of partially occupied water molecules is pretty
tough, particularly when the ligand is, itself, superimposed on
partially occupied water.

   To keep my focus on some sort of an absolute scale I always think
of contour levels in terms of electrons/A^3.  This calculation is also
problematic as is described in Lang et al.  Another tool I use is to
leave out a fully occupied water molecule that I have confidence in.
I can compare mystery density with this known control to see if I am
puzzling over a fully occupied atom or something weaker.  With a raw
Fo-Fc map you can never be sure.

   This "1 and 3 sigma" business really needs to be stomped out.

Dale Tronrud

> 
> On Wed, Sep 10, 2014 at 9:26 AM, Pavel Afonine <pafonine at lbl.gov>
> wrote:
>> Hi Charles,
>> 
>> I think the key here is what you call "no difference density
>> peak". Sites may be partially occupied so rule of thumb for
>> choosing contouring levels for 2mFo-DFc and mFo-DFc maps (1 and 3
>> sigma, correspondingly) may not be appropriate, for instance.
>> 
>> Pavel
>> 
>> 
>> On 9/10/14 6:10 AM, CPMAS Chen wrote:
>> 
>> Thanks, Nat.
>> 
>> If this is noise, why the anomalous or LLG peaks could be as high
>> as 5 ~ 6 sigma?
>> 
>> Charles
>> 
>> On Tue, Sep 9, 2014 at 5:58 PM, Nathaniel Echols
>> <nechols at lbl.gov> wrote:
>>> 
>>> On Tue, Sep 9, 2014 at 8:40 AM, CPMAS Chen <cpmasmit at gmail.com>
>>> wrote:
>>>> 
>>>> Is it possible that I have anomalous and LLG peak, but I have
>>>> no difference density peak? In this case, is that because the
>>>> model I have is not good enough or the diffraction data at
>>>> this site of the model is missing?
>>> 
>>> 
>>> You should at a minimum see > 1sigma density in the 2mFo-DFc
>>> map, and if the site is unmodeled (or modeled as a water) you
>>> should see an mFo-DFc peak as well.  If neither of these
>>> applies, the anomalous and LLG peaks are probably just noise.
>>> 
>>> -Nat
>> 
>> 
>> 
>> 
>> --
>> 
>> ***************************************************
>> 
>> Charles Chen
>> 
>> Research Associate
>> 
>> University of Pittsburgh School of Medicine
>> 
>> Department of Anesthesiology
>> 
>> ******************************************************
>> 
>> 
>> 
>> _______________________________________________ phenixbb mailing
>> list phenixbb at phenix-online.org 
>> http://phenix-online.org/mailman/listinfo/phenixbb
>> 
> 
> 
> 
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (MingW32)

iEYEARECAAYFAlQ4HQsACgkQU5C0gGfAG11ycwCfVX0p/LceJbe6mDvVUwdXC1Gn
/J4AoKzKrMjJJ7UlkaPlNjSlij6aOoIp
=3upb
-----END PGP SIGNATURE-----


More information about the phenixbb mailing list