[ccp4bb]: FWD: Contours of a discussion (SUMMARY; LONG !)

[Alun Ashton <A.W.Ashton@dl.ac.uk>]

***  For details on how to be removed from this list visit the  ***
***          CCP4 home page http://www.ccp4.ac.uk         ***


Dear All,

Gerard posted a follow up message on the 8th of Sept on the contours
'discussion'. However, due to the word 'su*****be' appearing in the first
few lines the bulletin board admin software here interpreted the message
to be a bb admin request.

I have included the message below and my apologies to Gerard and the bb
for the delay.

Alun
PS yes OK I will write a 'Wheres my message gone?'.html later for all you
others whos messages dont appear on the bb....
_____________________________________________________
Alun Ashton,   awa@ccp4.ac.uk    Tel: +44 1925 603528
CCP4,          ccp4@ccp4.ac.uk   Fax: +44 1925 603825
                      http://www.ccp4.ac.uk/
Daresbury Lab,  Daresbury,  Warrington,  UK,  WA4 4AD

---------- Forwarded message ----------
Date: Sat, 8 Sep 2001 01:14:29 +0100 (BST)

            *** SURGEON GENERAL'S HEALTH WARNING ***

If you subscribe to the CCP4 mailing list, but find crystallographic
discussions "boring", "annoying", "pedantic", or "off-topic", delete
this message now !

----------------------------------------------------------------------------

Right - that got rid of Carlo, then.

Well, what can I say ? Never a boring moment when Gerard (K) opens his (big)
mouth.

Let's try and retrace the discussion to see what (or: if) we can learn from
it.

It all started on Wednesday, 5 September, when Gosu Ramachandraiah asked a
simple question:

>Given a effective resolution of the data, at what contour one has
>to examine the mixed fourier synthesis map (2Fo-Fc & Fo-Fc) ?
>Is there any relation between resolution, completeness & contour ?

Bart Hazes was the first to reply:

>The simple answer is that you want to contour at a level that gives the
>clearest view of the density. I normally contour 2Fo-Fc & Fo-Fc at 1 and 3
>sigma, respectively, but in some parts of the map you want to lower the
>contour level. Your density will most likely not be equally strong throughout
>the map due to variations in B-factor or missing strong low resolution terms,
>so you'll have to adjust contour levels. If you lower the contour level too
>much you'll be blinded by all the noise features. Just use your eyes to tell
>what works best.

Wise words, except that the mentioning of a 3-sigma level in an Fo-Fc map
reminded me of the old days, when crystallographers still described the
details of their structure determinations, and when many people seemed to
blindly use the "3-sigma mantra" to justify their water molecules (those of
you who attended the 1996 Gordon Conference may remember my ranting about this
then - I think I called it "getting waters from the 3-Sigma catalogue").

So I followed up Bart's posting with some remarks:

>- the sigma level of an Fo-Fc is meaningless. In the early
>  stages (poor and incomplete model), a 2-sigma feature
>  may be genuine, whereas near the end of the refinement
>  process (when the difference map is hopefully flat
>  except for noise) even a 5-sigma peak need not be

This was my first remark, and the one that caused a lot of confusion
obviously. Of course, my formulation was careless: the sigma level (in
absolute terms) is not meaningless. And it was stupid of me to use hyperbole
and write "5-sigma" instead of "3-sigma". I didn't understand what Andrew was
subsequently on about until I got Gerard B's mail. Of course, my point was not
at all about the number 5 (or even 3), it was about the fact that a sigma
level of an Fo-Fc map is not an absolute quantity by any means. Hence, you
cannot say that one should contour such maps at a level of so-many-sigma - the
number "so-many" should vary as refinement progresses, the model gets more
complete and better, and the difference map gets flatter and flatter. In
addition, as bart pointed out, one may need to adjust the level in different
parts of the model.

Note for the pedantics: I clearly wrote that a 5-sigma peak NEED NOT BE
SIGNIFICANT at the end of a refinement, which is not the same as saying that
such a peak is INSIGNIFICANT. So there !

My third remark is perhaps even more relevant to the discussion than the
first:

>- you want to be careful with going to too low a contour
>  level. For an example of the dire consequences that
>  can have, see Nature Structural Biology 8 (8), pp.
>  663-664 ... [If you need convincing, check the real-space
>  fit and the map for 1F83 (chain B and C) at the Uppsala
>  Electron-Density Server, http://portray.bmc.uu.se/eds ]

Students: read that discussion, read the original paper, examine the model,
the temperature factors, the map, and the real-space fit values. Then form
your own judgment.

Andrew replied thusly:

>I seriously disagree with this remark. The sigma level in an Fo-Fc
>map does indeed give an indication of the noise level of the map, and
>one would certainly hope that this noise level, in absolute terms,
>would go down as refinement proceeds.

Which was exactly my point to begin with.

>However, this certainly does NOT
>mean that a 5 sigma peak in a difference map near the end of
>refinement is not suggesting an error in the model in that region.

Nor did I ever say it was. And had I said "3-sigma" Andrew might not have
replied at all ;-)

>Whether or not one can interpret the feature is a different matter, as
>very often they arise from multiple conformations/disorder that is
>very difficult if not impossible to model correctly.

Freshly out of bed, I found Andrew's posting and failed to understand what he
disagreed with:

>i seriously disagree with your seriously disagreeing.
>assume that the sigma level at the end of a refinement
>is 0.0001 e/A3. then a 10-sigma feature is noise. my point

Again - silly of me to use hyperbole, because this parenthetic,
besides-the-point remark of my got Andrew to react again. Whereas it was the
following bit that was my actual point (and also Andrew's original point,
mutatis mutandis, and my original point, if unclearly formulated ;-):

>is that the "number of sigmas" of a feature in an fo-fc
>map is meaningless - the absolute sigma level is not, of
>course (but this assumes that you calculate your maps on
>an absolute scale to begin with !).

Andrew retorted:

>I would suggest that your hypothetical example is not of any relevance
>to macromecular crystallography.

Absolutely.

>There are ALWAYS errors in a model,

He's telling *me* :-)

>even in the final stages of refinement, because there is never
>sufficent data to be able to model all aspects of conformational
>mobility/disorder.

I've never claimed otherwise.

>A 5 sigma peak in an Fo-Fc difference map, even
>close to the "end" of refinement, can be interpreted with very high
>confidence as indicating an error in the model. As I mentioned
>earlier, in many cases it may not be possible to interpret these
>features, but this does not mean they have no significance.

Which I never said in the first place. And again it's all about the number "5"
- which I still didn't understand. Feeling incapable of conveying my points, I
went back to my original ones and posted those again. Just after that I got
Gerard B's mail and only then did it dawn upon me that all of this was
precipitated by the number "5".

At this stage I was had enough of repeating myself, so I stopped reading mails
with "contour" in the subject line (until just now) and did some real work for
a change. (Amongst other things I'm involved in the organisation of a workshop
here in Uppsala in which Andrew will be the main attraction :-)

Meanwhile on the CCP4 bulletin board, Jeff Taylor took the discussion back to
the original question:

>So if I follow, that annoying 5 sigma peak in my difference map is
>either meaningless, or significant but not interpretable?

Phil Jeffrey also addressed the 5-vs-3 issue and gave some practical advice
that everyone should be able to agree with:

>Although I would hate to have to retreat to wearing sack-cloth and
>speaking Latin, it's not unreasonable to contour Fo-Fc maps at plus and
>minus 3 sigma for the purpose of finding things to look at.  One will want
>to consider things like geometry, environment, and yea even electrons per
>cubic Angstrom in assessing the peak, but dogma is no substitute for care
>and experience in interepreting such features.  It might work quite well
>for buying beer, however.

(At this stage, Carlo got fed up with the discussion.)

Andrew then took my remarks about waters from the 3-Sigma catalogue personal,
and retaliated by misinterpreting my words (see the "note for pedantics"
above).

At this time I started receiving a number of mails that were not copied to
CCP4BB as well - many of them saying that a bit of discussion on the mailing
list is good, especially for younger crystallographers.

Mike Sierk wrote something similar, but to the mailing list:

>I disagree.  I think these kinds of discussions (flame wars?) are
>valuble to younger researchers like myself, both because they go over
>relevant technical details, and because they point up the fact that the
>science is not always clear-cut and different people do (and interpret)
>things in different ways.
>
>If you aren't interested, or if the discussion gets overly pedantic,
>that's what the Delete button is for : )

Bart Hazes realised what had been happening earlier:

>I think in practice Andrew and Gerard are doing the same thing.

And he went on to list further sensible ways of dealing with the problem of
adding waters to a model.

So, back to the original question. At what level should one contour a
difference map ? Well, one trick that may be useful is to leave out a
well-defined atom (e.g. a carbonyl oxygen) in the map calculation and adjust
the Fo-Fc contour level until that density looks just as good as the 2Fo-Fc
density for the same (missing) atom. Then you know that well-ordered entities
with ten-or-so electrons should have similar density features in both maps.
This is completely general. When it comes to water molecules in particular,
obviously one should use other criteria as well (plausible hydrogen- bonding
partners, refine to reasonable B-factors, possess acceptable 2Fo-Fc density
after refinement).

By the way - many of these issues are discussed in chapter 11 of Methods in
Enzymology, volume 277, pp. 208-230 (1997) ...

Finally, in my experience three very useful tools in macromolecular model
building are:

         (1) Ockham's razor (don't cut yourself !)
         (2) Common sense (a cheap but neglected asset)
         (3) CD-Walkman (with a Ramones CD)

And I will try to cut back on the use of hyperbole in my postings. In fact,
from now on, I will proof-read my postings at least a
hundred-million-gazillion times before hitting the "send" button ...

Yes, I will shut up now.

--Gerard

By the way: last week I prepared a web-based practical about model validation
for students in the EMBO Bioinformatics Course we had here in Uppsala.
Although this practical was aimed at non-crystallographers, it may be useful
for relative newcomers to the area as well. It can be found at:

    http://xray.bmc.uu.se/embo2001/modval


******************************************************************
                        Gerard J.  Kleywegt
    [Research Fellow of the Royal  Swedish Academy of Sciences]
Dept. of Cell & Molecular Biology  University of Uppsala
                Biomedical Centre  Box 596
                SE-751 24 Uppsala  SWEDEN

    http://xray.bmc.uu.se/gerard/  mailto:gerard@xray.bmc.uu.se
******************************************************************
   The opinions in this message are fictional.  Any similarity
   to actual opinions, living or dead, is purely coincidental.
******************************************************************