[3dem] pdb for near-atomic resolution EM maps

Petr Leiman petr.leiman at epfl.ch
Tue Dec 2 02:12:41 PST 2014

Dear Ariel,

I think there is a major confusion about what is modeling and what are experimentally determined constrains. Rotamer libraries are derived from experimental data, just like the rest of geometrical parameters. Low resolution refinement and data interpretation for that matter heavily rely on rotamer libraries. I mentioned MD in my previous message as something that seriously dedicated people might want to try using the structure that has been deposited to PDB or elsewhere as a main chain for which side chains have just been recreated. I never suggested depositing MD simulated data to PDB.

In principle, side chains at 4-5... A resolution are similar to hydrogen atoms at ~1-1.2 A in X-ray. You can't really see most of them, but you know they are there and they can be easily built using geometry (and the crystallographic community has annual discussions of what to do with H atoms - deposit or not). Conformational space of most side chains is much greater than that of hydrogens, but rotamer libraries help us to reduce it to just a few conformations and this is what ALL the refinement programs use at low resolution.

On the other hand, I am a little disturbed by your remark that you chose to deposit a structure that clearly does not agree with your experimental map. Shouldn't it be a C-alpha trace or something similarly less precise than a complete atomic model?

Best wishes,


On 12/02/2014 10:38 AM, Ariel Blocker wrote:
Dear Irina,
I agree with the other replies that this is a key question for the EM community now.
We recently had a similar "cas de conscience”, but with a crystal structure docking to an even much lower resolution map, where it is absolutely obvious that none of the side chains can be accurate (and maybe the polypeptide chain fold isn’t vaguely right in places!). We decided to deposit the pseudoatomic model in the pdb anyway, partly because it is what EMDB recommend, partly because our journal editor also said he felt it was appropriate but also partly because we felt that if we didn’t nobody could independently reassess/test/use our work in the future. Now, in theory, all those accessing such files should be able to figure out for themselves how accurate the models are but, in reality, many I feared would be fooled into taking their visual accuracy for granted. So, we decided to put a “resolution health warning” as a note in the deposition, which I requested would be available upfront to the downloaders. And, of course, we also discussed the crudeness of the resolution in the paper.
Regarding the rapidly increasing numbers of 4-5 Å resolution cryoEM maps, I think we should also caution reviewers and users about relying on the rotamer libraries and MD (rather than additional ED map refinement, where possible, which should be done by authors) to improve the pseudoatmoic models. That is mixing modelling with experimental data in ways that, as far as I know, are not properly experimentally investigated/validated yet (but evidently need to be in future), and where there is therefore again serious concern about overinterpretation. So, if we chose do it, it should also be explicit in the deposition and paper.

Ariel J. Blocker, PhD, FSB
Reader in Microbiology
Wellcome Trust Investigator
Schools of Cellular & Molecular Medicine and Biochemistry
Medical Sciences Building
University of Bristol
University Walk
Bristol  BS8 1TD
United Kingdom

Email: ariel.blocker at bristol.ac.uk<mailto:ariel.blocker at bristol.ac.uk>
Tel office (D39a): +44 117 33 12063
Tel wetlab (D57/D59): +44 117 33 12404
Tel wetlab office (D53): +44 117 33 12059
Tel drylab (D51): +44 117 33 12097
Tel Tecnai20: +44 117 33 12357
Dept. fax: +44 117 3312091
Room: D39a
Personal webpage: http://www.bristol.ac.uk/cellmolmed/research/infect-immune/blocker.html
Group page: http://www.bristol.ac.uk/cellmolmed/research/infect-immune/blocker/
Postgraduate students searching for a PhD please look here:

On 2 Dec 2014, at 09:03, Petr Leiman <petr.leiman at epfl.ch<mailto:petr.leiman at epfl.ch>> wrote:

On 12/02/2014 08:43 AM, Irina Gutsche wrote:

Dear all,

I would like to deposit a near-atomic resolution map (4.3A) and the
corresponding atomic model to the respective databases. Given recent
discussions about overinterpretation of the data I am wondering what
would be the best way to proceed.
Indeed, some bulky side chains are clearly visible and could be placed
with confidence, in particular in certain regions of the map
(according to ResMap local resolution ranges from 3.5A to above 6A),
but most of them not. And even the criteria of "visibility" and
"confidence" are subjective. The protein complex is of big
pharmaceutical interest and I suppose that if the complete pdb with
all the side chains is deposited, it may be downloaded and used for
example for drug design, as a kind of "absolute truth", which might be
dangerous. However, deposit only the c-alpha is probably too
restrictive. Another option would be to deposit only the side chains
we are "sure about", but again, it's subjective, and anyway we can
only be "sure" at the given resoluition. So what would be the most
honest and unbiased way to share these results with the scientific
community, in particular with biologists who are not necessarily aware
of the overinterpretation/overfitting/validation/etc issues in our
field but who use our results to design their experiments?

This kind of topic is discussed at crystallographic BBs at length about
once a year for as long as those boards exit. There, however, people
discuss what to do with side chains for which there is no density. Build
them or truncate at C-alpha or C-beta or make them zero occupancy or let
the refinement take care of the problem (it will elevate the B factor to
hundred(s) of A^2). These discussions are endless for the exact same
reasons that Irina has brought up.

In the case of 3.5-6 A resolution structure, the experimental error of
atomic positions is so high that you build your side chains based on the
rotamer database, but not on the density. There is also an uncertainty
(e.g. error) associated with placing the main chain, but let's forget
about this for a second. My point: for these "almost atomic
interpretation" datasets, the information contained in the main chain
that is placed into the density to the best of our ability is all that
is required to recreate the complete atomic model. For simple
interpretations and ribbon diagrams, the main chain information is
sufficient. But people who want to design drugs will have to look into
the model a little more. They will have to rebuild all the side chains
(they will rely on the rotamer library for doing so - exactly what we do
now) and perform MD and/or additional refinement. And of course, the
main chain model should contain all the side chains for which there is
good experimental density.

Best wishes,


Thank you very much for your all ideas and suggestions,

Best regards,

Irina Gutsche
CNRS, Grenoble,

3dem mailing list
3dem at ncmir.ucsd.edu<mailto:3dem at ncmir.ucsd.edu>

Petr Leiman
Laboratory of Structural Biology and Biophysics

CH-1015 Lausanne
Phone: +41 21 69 30441
Mobile: +41 79 538 7647
3dem mailing list
3dem at ncmir.ucsd.edu<mailto:3dem at ncmir.ucsd.edu>

3dem mailing list
3dem at ncmir.ucsd.edu<mailto:3dem at ncmir.ucsd.edu>

Petr Leiman
Laboratory of Structural Biology and Biophysics

CH-1015 Lausanne
Phone: +41 21 69 30441
Mobile: +41 79 538 7647

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.ncmir.ucsd.edu/pipermail/3dem/attachments/20141202/66890c62/attachment-0001.html>

More information about the 3dem mailing list