Dear Thierry,

     This is a very interesting approach and observation. Using a
threshold of 0.6 rmsd, if converted into a ratio of B-factors, would 
mean accepting waters that have a 40% higher B-factor than would be
done with a 1.0 rmsd threshold.

     Philippe Dumas commented off-list on my anecdote regarding the
extra copy of a molecule only becoming convincingly visible at 0.35
rmsd, and he pointed out that it could have been a case of partial
occupancy rather than of a uniformly higher B-factor. That is a very
good point, which we haven't pursued. However the conclusion is the
same: the contouring level at which a density map is examined must be
adapted to the model of disorder for that region that one is prepared
to consider: for an increased B-factor, using the formula I outlined;
or for a partial occupancy, using a corresponding fraction of the
contour level at which full-occupancy regions are examined. The main
thing is that the contour level in rmsd units isn't the same thing as
a signal-to-noise ratio.


     With best wishes,
     
          Gerard.

--
On Wed, May 27, 2015 at 01:01:50PM -0400, Fischmann, Thierry wrote:
> Here is another observation regarding the subject of which rms value is more sensible to review a structure.
> 
> The purpose of the calculation described below was to determine the optimum e- density value for rejecting waters.
>  
> The principle of the test was very simple : waters were added using the "traditional" peakmax / H-bond distance based criteria. The cut-off in difference peaks for water selection was deliberately set to be "low" (I forgot what it was, as it's been a while since I've performed these tests). The structures were refined, then waters were rejected if the sigmaa-weighted 2Fobs-2Fcalc density value was less than a specific threshold. A contact check was also used for water rejections. The structure was then given another final round of positional refinement. All refinement were performed using BUSTER. The Rfree values were then compared : the optimum e- density cut-off value would be - obviously - the one for which the value of Rfree is lowest.
> 
> The tests were performed with a few datasets, all good resolutions although I can't remember what they were (definitely better than 2Å). One caveat was that these tests were performed only for one project i.e. human Cdk2 kinase. A consistent set of "free" reflections between datasets and the one from the PDB used as starting point was used, as you would expect.
> 
> The best cut-off according to the Rfree criterial was consistently about 0.6 rmsd. It most likely could vary depending on the crystal system and the resolution. But in the simple scenario described above the "best" value was quite a bit lower than the "traditional" cut-off of 1.0
> 
> 
> Thierry
> 
> -----Original Message-----
> From: CCP4 bulletin board [mailto:[log in to unmask]] On Behalf Of Gerard Bricogne
> Sent: Wednesday, May 27, 2015 8:32 AM
> To: [log in to unmask]
> Subject: Re: [ccp4bb] Validity of Modeling of Highly flexible region at low sigma levels
> 
> Dear Sagar and Pavel,
> 
>      Yes, it may be model bias, but you also have to bear in mind that
> the rmsd that is used as a unit in the choice of contouring level
> should not be thought of as the standard deviation of a Gaussian
> random variable and therefore implicitly associated to a "significance
> level" relative to a noise level. This is unfortunately a widespread
> misconception.
> 
>      We have seen a case where an entire copy of a molecule had been
> missed at first because its density was visible only at a contour
> level of 0.35 "sigma", which at first sight would make anybody shrink
> back in horror. When it was nevertheless modelled and refined, there
> was no doubt whatsoever that it was there, making perfectly sensible
> contacts with its nearest neighbour molecules, obeying good NCS with
> the already placed copies and markedly improving the fit to the data.
> The peculiarity of that extra copy was that a clash around a symmetry
> element with its symmetry mate caused it to have a mean B-factor that
> was twice that of the other, well-ordered, copies of the molecule.
> 
>      Now, if you look at the expression for the Debye-Waller factor in
> real space, you will find the B-factor raised to the power 3/2 in the
> denominator. That means that the blurring effect of that D-W factor in
> real space results in a lowering of the maximum value of an atomic
> electron density to which it is applied by B^(3/2). 
> 
>      The consequence is that electron density for a molecule whose
> mean B-factor is double that of another molecule will look similar at
> a contour level of 1.0/(2^(3/2)) = 0.35 to what the other molecule
> looks like when examined at a contour level of 1.0.
> 
>      It is therefore a good idea to remember that the "sigma" level is
> mainly determined by the better-ordered region(s) of the molecule(s)
> present in the asymmetric units, and that less-well ordered regions
> will automatically have a contouring level handicap that is not to be
> automatically interpreted as a lower level of significance. Omit maps
> are a good idea of course, but do bear in mind the (B/B0)^(-3/2)
> effect, where B0 is the mean B in the well-ordered region, and B is
> the same quantity in a less well ordered one. Above all, don't confuse
> "sigma" with a noise level!
> 
> 
>      With best wishes,
>      
>           Gerard.
> 
> --
> On Wed, May 27, 2015 at 12:18:25PM +0100, Pavel Afonine wrote:
> > Sagar,
> > 
> > what you see may be model bias unless you calculated these maps without
> > that region of molecule.
> > 
> > Pavel
> > 
> > On Wed, May 27, 2015 at 12:03 PM, Sagar De'Biomimic <
> > [log in to unmask]> wrote:
> > 
> > > Dear all,
> > >
> > > We have solved a structure of Protein-DNA complex. In an ASU we have two
> > > protein homodimers and two DNA duplexes. Additionally we were able to model
> > > N-terminal region (NTR) missing in the previously reported structure.
> > >
> > > I had modelled poly-alanine chains in the positive density and refined.
> > > These chains were then joined to form NTR. Finally the entire NTR was
> > > refined for occupancies.
> > >
> > > I am doubtful about the validity of modelling of NTR as i had dropped down
> > > sigma levels to as low as 0.8 (as i could not see much of the noise) of
> > > FO-FC and modelled in positive density. After refinment the region shows
> > > the density at 0.5 sigma level of 2FO-FC. I have attached a pdf file
> > > showing snapshots of NTR at 0.4, 0.5, 0.6, 0.7 sigma level of 2FO-FC map.
> > >
> > > I would like to know crystallography community's opinion on validity of
> > > such modelling.
> > >
> > > This NTR region is highly flexible. Main intention of modelling it in such
> > > a weak signal was to complete the model for our Molecular dynamics
> > > studies.
> > >
> > > Additionally we have confirmed with SAXS DATA using Ensemble optimised
> > > modelling (EOM 2.0), the high flexibility and multiple conformations of
> > > this NTR region.
> > >
> > > I would like to know if there is a possibility of having a tool similar to
> > > EOM to model highly flexible regions of a protein in electron Density. As
> > > far as I know xMDFF does better at lower resolutions but fails to model
> > > highly flexible regions such as ours.
> > >
> > > Thank you.
> > >
> > > Regards
> > >
> > > Sagar
> > >
> > >
> > >
> > > --
> > > Sagar Khavnekar
> > > Project student,
> > > Structural Biology and Molecular Biophysics lab,
> > > UM-DAE Centre for Excellence in Basic Sciences
> > > University of Mumbai, Vidyanagari Campus, Kalina, Santacruz (East)
> > > Mumbai 400098, India.
> > >
> 
> -- 
> 
>      ===============================================================
>      *                                                             *
>      * Gerard Bricogne                     [log in to unmask]  *
>      *                                                             *
>      * Global Phasing Ltd.                                         *
>      * Sheraton House, Castle Park         Tel: +44-(0)1223-353033 *
>      * Cambridge CB3 0AX, UK               Fax: +44-(0)1223-366889 *
>      *                                                             *
>      ===============================================================
> Notice:  This e-mail message, together with any attachments, contains
> information of Merck & Co., Inc. (2000 Galloping Hill Road, Kenilworth,
> New Jersey, USA 07033), and/or its affiliates Direct contact information
> for affiliates is available at 
> http://www.merck.com/contact/contacts.html) that may be confidential,
> proprietary copyrighted and/or legally privileged. It is intended solely
> for the use of the individual or entity named on this message. If you are
> not the intended recipient, and have received this message in error,
> please notify us immediately by reply e-mail and then delete it from 
> your system.