[3dem] [ccpem] Relion 3.1 CtfRefine anisotropic magnification

[Marin van Heel]

Hi Omid,

We recently had a very complicated case of reference bias: using a
rotationally-symmetric reference image to pick spherical viruses sounds a
logical thing to do does'nt it? However, it actually caused the data set to
split into two: We got two 3D reconstructions differing in size by 0.9%.
It took me more than a week to understand that that disparity was caused
primarily by the picking of elliptical particles (2.6% anisotropic
magnification) using a rotationally-symmetric reference. It was a very
tricky case of reference bias...  So don't expect me to explain the nitty
gritty details in less than 10 lines of email. Bottom line: our correction
algorithm (see paper) took fully care of it.  The underlying problem was
that we had not seriously looked at anisotropy due to an entirely different
issue that obscured the data: our Cs-corrected Krios suffered from wild
unpredictable astigmatism fluctuations during data collection (up to 1
micron!), for almost three months. Only after that issue was resolved did I
even try to measure anisotropic magnification.
Hope this helps,

Marin


On Tue, Jan 21, 2020 at 7:19 PM Haji-Ghassemi, Omid <omidgh85 at mail.ubc.ca>
wrote:

> Dear Marin,
>
>
>
> Hope your are doing well and thanks again for a memorable time in Brazil
> 😊. I apologize for the delay in response and I really appreciate your
> suggestion. It makes sense that anisotropic magnification is best performed
> on the initial movie stacks but I don’t think I completely comprehend how anistropic
> magnification correction lead to bias during picking if it is done after
> the initial processing (since the picking is now done anyway). Can you
> clarify?
>
>
>
> Best,
>
> Omid
>
>
>
> *From:* Marin van Heel <marin.vanheel at googlemail.com>
> *Sent:* January 17, 2020 8:54 AM
> *To:* Haji-Ghassemi, Omid <omidgh85 at mail.ubc.ca>; 3dem <
> 3dem at ncmir.ucsd.edu>; ccpem at jiscmail.ac.uk
> *Subject:* Re: [ccpem] Relion 3.1 CtfRefine anisotropic magnification
>
>
>
> Dear Omid,
>
>
>
> Anisotropic magnification should be corrected first, even before any CTF
> determination, since its presence messes up the defocus/astigmatism
> assessment. The anisotropic magnification measurement is best/conveniently
> performed on the movie data itself. It does not require any separate
> calibration measurements since the water ring information is always present
> in large cryo-EM datasets, and those contain all the necessary anisotropy
> information. The Imagic procedures allow you to a posteriori correct
> anistropic magnification even in "legacy" cryo-EM datasets. Depending on
> how you perform the correction, the average pixel size may change, and
> consistency is important. It is not a good idea to do this anisotropic
> magnification correction as an "afterthought" at the late stages of
> refinement since it may bias other intermediate processing steps like
> particle picking.
>
> We have described the procedure extensively in:
>
>  Afanasyev P, Seer-Linnemayr C, Ravelli RBG, Matadeen R, De Carlo S,
> Alewijnse B, Portugal RV, Pannu NS, Schatz M, van Heel M: *Single-particle
> cryo-EM using alignment by classification (ABC): the structure of Lumbricus
> terrestris hemoglobin*, *IUCrJ. **4* (2017) 678-694.
>
>
>
> Hope this helps,
>
>
>
> Marin van Heel
>
>
>
> On Tue, Dec 24, 2019 at 7:30 PM Haji-Ghassemi, Omid <omidgh85 at mail.ubc.ca>
> wrote:
>
> Dear All,
>
>
>
> I recently tried estimating the anisotropic magnification via CtfRefine in
> Relion 3.1 and I get the following warning once the run was finished:
>
> WARNING: Overall magnification of optics group #1 (opticsGroup1) differs
> from the nominal pixel size by 1.92044 %.
>
> WARNING: This overall difference changes the actual pixel size of the
> reconstruction!
>
>
>
> I also attached the resulting log pdf file.
>
> I then notice subsequent 3D refinement yielded strange 3D volume and
> really poor data quality. I have refined this dataset before and I obtain a
> resolution of ~-3.8-3.9 Ang. I also attempted to calibrate the pixel size
> on a region of the map where the resolution is the highest, and my pixel
> size seemed correct. The pixel size of the data is supposed to be around
> 1.08 or 1.09.
>
>
>
> So my questions are as follows:
>
> 1. Up to what resolutions can the anisotropic magnification be detected?
>
> 2. If indeed the data suffers from some degree of pixel error or
> magnification issues, do I need to adjust it manually before submitting the
> next refinement job? I assumed the correction is already applied after the
> CtfRefine job. If so where is this information recorded? The pixel size
> remained the same in the .star file of the header.
>
> 3. How do I analyze the log file data? It was not clear to me from after
> consulting the tutorial and the paper by Zivanov et al., 2019.
>
>
>
> Regards,
>
> Omid
>
> ---------------------------------------------------
>
> ---------------------------------------------------
>
> Omid Haji-Ghassemi, Ph.D.
>
> Postdoctoral Fellow
>
> Department of Biochemistry & Molecular Biology
>
> Faculty of Medicine
>
> University of British Columbia
>
> 2350 Health Sciences Mall
>
> Vancouver, BC, V6T 1Z3
>
> Canada
>
>
>
>
> ------------------------------
>
> To unsubscribe from the CCPEM list, click the following link:
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCPEM&A=1
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.ncmir.ucsd.edu/pipermail/3dem/attachments/20200122/bec49d1d/attachment-0001.html>