[3dem] 3dem Digest, Vol 79, Issue 11
Sjors Scheres
scheres at mrc-lmb.cam.ac.uk
Thu Mar 13 07:46:30 PDT 2014
Hi Reza,
Let me add our little bit to this discussion. We use the FSC=0.143
criterion based on RELION's "gold-standard FSCs" (sorry Marin) to
determine resolution and have observed the following:
resolution interpretation
--------------------------------------
>10-15A molecular details are difficult to interpret, mainly
limited to protein/domain assignment
~9A helices become sausages
8-5A everything becomes better in this range, but
improvements tend to be rather gradual
<5A bulky side chains start becoming visible
~4.7A beta-strands become separated
~4A de-novo building starts being OK, more side chains
appear in the density, refinement of the atomic model starts behaving
rather well
3.5A nucleotide bases start being separated, most of the
side chains are visible in the density
3.5-3.0A de-novo building becomes ever easier, refinement of the
atomic model in the map becomes very easy
In summary: 9, 4.7 and 3.5A are key resolutions. We haven't gone to 2A
yet. That would be awesome though! :-)
Also, we now always estimate local resolution variations in our maps, as
very few complexes are equally stable throughout. We have had very good
experiences using ResMap when using the 2 unfiltered, independent
half-reconstructions from RELION auto-refinements as input on the "Split
Volume Input" tab of ResMap.
Hope that helps,
Sjors
On 03/13/2014 11:02 AM, rkhayat at ccny.cuny.edu wrote:
> Hi,
>
> I would like to ask a much more naive question. Rather than asking how does
> one determine resolution, I would like to ask how does resolution alter our
> interpretation of an image reconstruction? For example, what more can we
> learn of a complex if we generate a correct image reconstruction of it at
> 30Angstroms vs 25, 20, 16, 12, 9, 7, 5, 3 and 2Angstroms. How does this
> interpretation change if we do not posses and do posses PDB(s) of the
> components? What and where are the resolution landmarks that signify an
> advance in interpreting an image reconstruction (e.g. subnanometer resolution
> allows helices to be identified)?
>
> Best wishes,
> Reza
>
> Reza Khayat, PhD
> Assistant Professor
> The City College of New York
> Department of Chemistry, MR-1135
> 160 Convent Avenue
> New York, NY 10031
> Tel. (212) 650-6070
>
>
> ---- Original message ----
>> Date: Wed, 12 Mar 2014 22:05:10 +0000
>> From: 3dem-bounces at ncmir.ucsd.edu (on behalf of Qiu-Xing Jiang <Qiu-
> Xing.Jiang at UTSouthwestern.edu>)
>> Subject: Re: [3dem] 3dem Digest, Vol 79, Issue 11
>> To: "3dem at ncmir.ucsd.edu" <3dem at ncmir.ucsd.edu>,"Sindelar, Charles"
> <charles.sindelar at yale.edu>
>> Hi Chuck, We did the FSC of the "gold-standard" calculations in the paper.
>> I was not sure about the PDB modeling in determining resolution in our
>> case, even though it was "better than nothing" as you said. Thanks for
>> sharing your thoughts. Qiu-Xing
>>
>>
>>
>> On 3/12/14 4:45 PM, "Sindelar, Charles" <charles.sindelar at yale.edu> wrote:
>>
>>> Hi Qiu-Xing, assuming the question you want to answer is "what is the
>>> highest resolution I can claim with certainty", then you have very few
>>> options. Really the best one is the FSC of "gold-standard" calculations
>>> where the two halves of the data don't see each other, as was already
>>> mentioned. As you pointed out, FSC between your map and rigid-docked
> PDB
>>> can potentially under-estimate the resolution, but it's better than
>>> nothing, because it delivers a guaranteed lower limit.
>>>
>>> On the other hand, if your question is "how high could my resolution be"
>>> - then you are free to use any method you like, and pick the highest
>>> number. But obviously there are shortcomings with this approach.
>>>
>>> - Chuck
>>>
>>> On Mar 12, 2014, at 5:27 PM, Qiu-Xing Jiang
>>> <Qiu-Xing.Jiang at utsouthwestern.edu>
>>> wrote:
>>>
>>>> Hi Charles. I was asking whether a rigidly-docked PBD model is good
>>>> enough
>>>> to determine resolution for the membrane-facilitated filament because I
>>>> did't have a crystal structure of the filament itself.
>>>>
>>>> On 3/12/14 4:07 PM, "Sindelar, Charles" <charles.sindelar at yale.edu>
>>>> wrote:
>>>>
>>>>> Hi Qiu-Xing, the experiment that Ed refers to is one of the better ones
>>>>> you can do- the FSC between the PDB-generated map, and your
>>>>> experimental
>>>>> map. Particularly if you do only rigid-body fitting (not flexible
>>>>> fitting), the 0.5 cut-off will give you a conservative estimate of the
>>>>> resolution- i.e. it is difficult or impossible to over-estimate the
>>>>> resolution in this way. I was curious why your original email did not
>>>>> report the results of this type of experiment (forcing Ed to enlighten
>>>>> us). The other methods you used can all significantly overestimate the
>>>>> resolution of your map - including, notably, the ResMap program.
>>>>>
>>>>> On Mar 12, 2014, at 4:33 PM, <3dem-request at ncmir.ucsd.edu>
>>>>> wrote:
>>>>>
>>>>>> -----------------------------------------------------------
> -----------
>>>>>> Message: 1
>>>>>> Date: Wed, 12 Mar 2014 15:59:30 -0400
>>>>>> From: Edward Egelman <egelman at virginia.edu>
>>>>>> To: 3dem at ncmir.ucsd.edu
>>>>>> Subject: Re: [3dem] Resolution estimate: FSC vs PDB modeling
>>>>>> Message-ID: <5320BCA2.3050703 at virginia.edu>
>>>>>> Content-Type: text/plain; charset="iso-8859-1"; Format="flowed"
>>>>>>
>>>>>> I feel compelled to respond to this, as I am the "senior colleague".
>>>>>> The
>>>>>> paper under discussion appeared recently in Nature (2014), Mukherjee
>>>>>> et
>>>>>> al. Let me correct the record. If one compares the map (EMDB-5795)
>>>>>> with
>>>>>> the atomic model (4MTH) all correlation disappears at better than
>>>>>> about
>>>>>> 16 Angstroms, so the statement that there was strong correlation at
>>>>>> 9.2
>>>>>> to 15 Angstroms is wrong. I urge others to make this comparison
>>>>>> themselves. This led me to raise questions about the 9.2 Angstrom
>>>>>> resolution claim for the map. As far as I can see, after more than 20
>>>>>> emails exchanged with Dr. Jiang before the submission of the paper,
>>>>>> this
>>>>>> resolution claim was simply fraudulent. That was why I asked for my
>>>>>> name
>>>>>> to be removed from this paper! I would welcome, however, for the
>>>>>> cryo-EM
>>>>>> field to move to a "reality-based" standard of resolution, and this
>>>>>> paper provides a good basis for discussing why such a standard is
>>>>>> needed.
>>>>>> Regards,
>>>>>> Ed Egelman
>>>>>>
>>>>>> On 3/12/14, 8:47 AM, Qiu-Xing Jiang wrote:
>>>>>>> Dear colleagues,
>>>>>>> This has been a topic discussed in a couple of map validation
> papers.
>>>>>>> I have a scenario encountered in a recent project and would like
> some
>>>>>>> input from those interested.
>>>>>>> We had a filament complex reconstruction, made of a small protein
>>>>>>> whose crystal structure contains mainly loops held together in the
>>>>>>> core by three pairs of disulfide bonds. At the time of resolution
>>>>>>> estimation, we used two independent maps calculated from either
>>>>>>> randomly selected halves or top/bottom halves to calculate FSC as
>>>>>>> usual in SPIDER. The FSC0.5 of the former gives 9.2 angstroms,
> which
>>>>>>> is more conservative than the FSC0.143 of the latter, and was used
> as
>>>>>>> a nominal estimate.
>>>>>>>
>>>>>>> The other opinion from a senior colleague was PDB modeling, whose
>>>>>>> operations are to dock the X-ray structures of individual units into
>>>>>>> the map (assuming no change), filter the resulted PDB model to
>>>>>>> different resolutions, and visually determine at which resolution the
>>>>>>> map calculated from the PDB model and the experimental density
> map
>>>>>>> match the best(at certain thresholds), and could then be used for
>>>>>>> resolution estimate. When we were operating this procedure, we
> knew
>>>>>>> that our filaments had lipids associated and our map was not good
>>>>>>> enough to resolve the loops on the surface of the individual units in
>>>>>>> the filament. To make the two maps match well to our eyes, we had
> to
>>>>>>> filter both to 12-15 angstroms, which would then say the resolution
>>>>>>> was 12-15 A. The calculated cross-correlation between the map
> from
>>>>>>> the
>>>>>>> pdb model and the experimental map was high at 9.2-15 range, but
> we
>>>>>>> were not sure whether it would be decisively meaningful.
>>>>>>>
>>>>>>> We had debates and disagreements on which to report. At the end
> we
>>>>>>> decided to use FSC0.5 as usual, and refrained from interpreting the
>>>>>>> map with mutations at the subunit interface due to the experience of
>>>>>>> PDB modeling. I wonder if some of you had similar experience, and
>>>>>>> more generally whether PDB modeling is suitable to replace FSC.
>>>>>>>
>>>>>>> Thanks for sharing your experience.
>>>>>>>
>>>>>>> Best regards,
>>>>>>>
>>>>>>> Qiu-Xing
>>>>>>>
>>>>> Charles V. Sindelar, Ph.D.
>>>>> Dept. of Molecular Biophysics and Biochemistry
>>>>> Yale University
>>>>> SHMC-E25
>>>>> 333 Cedar Street
>>>>> New Haven, CT 06520-8024
>>>>>
>>>>> Phone (203) 737-4752
>>>>> Lab (203) 737-4723
>>>>> Fax (203) 785-7979
>>>>> http://medicine.yale.edu/mbb/faculty/charles_sindelar.profile
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> 3dem mailing list
>>>>> 3dem at ncmir.ucsd.edu
>>>>> https://mail.ncmir.ucsd.edu/mailman/listinfo/3dem
>>>>
>>>> ________________________________
>>>>
>>>> UT Southwestern Medical Center
>>>> The future of medicine, today.
>>>>
>>> --
>>> Charles V. Sindelar, Ph.D.
>>> Dept. of Molecular Biophysics and Biochemistry
>>> Yale University
>>> SHMC-E25
>>> 333 Cedar Street
>>> New Haven, CT 06520-8024
>>>
>>> Phone (203) 737-4752
>>> Lab (203) 737-4723
>>> Fax (203) 785-7979
>>> http://medicine.yale.edu/mbb/faculty/charles_sindelar.profile
>>>
>>>
>>>
>>>
>>
>> _______________________________________________
>> 3dem mailing list
>> 3dem at ncmir.ucsd.edu
>> https://mail.ncmir.ucsd.edu/mailman/listinfo/3dem
> _______________________________________________
> 3dem mailing list
> 3dem at ncmir.ucsd.edu
> https://mail.ncmir.ucsd.edu/mailman/listinfo/3dem
--
Sjors Scheres
MRC Laboratory of Molecular Biology
Francis Crick Avenue, Cambridge Biomedical Campus
Cambridge CB2 0QH, U.K.
tel: +44 (0)1223 267061
http://www2.mrc-lmb.cam.ac.uk/groups/scheres
More information about the 3dem
mailing list