Hi Oli,
Yes, this is a problem that has always plagued EM. Unfortunately it was never really dealt with because it's not much of an issue for lower resolution image reconstructions unless you're doing very detailed difference maps from the X-ray model. The problem is more pronounced now because atoms are finally being modelled into the reconstructions. This is something that most microscopists are not familiar with and thus do no go through the rigour of pixel correction. Most magnification/pixel size calibrations involve using diffraction from a 2D crystal with known cell parameters. This, in my opinion, is insufficient given that protein isomorphous x-tals are not all that common. Moreover, we know from our x-ray crystallography experiments that the final lattice constants of a x-tal are calculated during the scaling stage not just the indexing stage.
I would imaging that the rmsd values for your bonds and angles are a bit messed up as well. Your fastest way of dealing with the situation is to write scripts that modify the pixel size and calculate CC-values or R/Rfree factors between your model and the image reconstruction.
Reza
Reza Khayat, PhD
Assistant Professor
City College of New York
Department of Chemistry
New York, NY 10031
________________________________
From: Oliver Clarke
Sent: Monday, February 8, 2016 7:36 AM
To: Reza Khayat
Cc: [email protected]
Subject: Re: [phenixbb] Refine pixel size of map (for EM data)?
Hi Reza,
Second question first - not much (I can't detect any), assuming same scope and mag (which is relieving for multiple reasons). However, I don't know how close the pixel size I obtain by real space correlation with a crystal structure is to the "true" pixel size, and whether it is wrong in any systematic way.
Re the other questions, I have not compared rigorously (I only refined against the map with the correct pixel size), but I undertook steps to correct it when I noticed during initial building in Coot that known (solved) domains were not quite fitting the density as well as expected. I would expect some consequences from this, particularly in terms of clash analysis and the use of reference restraints (not sure about electrostatics etc).
Where it does make a big difference is in comparing structures of the same or related proteins solved by different groups - if the pixel size varies by even 1-2%, this really messes up structural superposition for large structures (if the structure is 300Å, this could be a 6 Å difference in size). In my experience many groups don't seem to do this - there are quite a lot of recent maps in the EMDB in which the nominal pixel size has been taken as correct, and calibration with a crystal structure shows that it is a ways off (this also affects the estimated resolution, sometimes substantially).
Cheers,
Oli
On Mon, Feb 8, 2016 at 11:38 AM, Reza Khayat mailto:[email protected]> wrote:
Hi Oliver,
Out of curiosity:
1. what kind of R-factors and CC-values do you get when refining against the two different pixel size?
2. how different are your refined pixel sizes from one reconstruction to another?
3. how much of an affect does the wrong pixel size have on your downstream structure analysis (e.g. BDA, ASA, electrostatic...)?
Best wishes,
Reza
Reza Khayat, PhD
Assistant Professor
City College of New York
Department of Chemistry
New York, NY 10031
________________________________
From: [email protected]mailto:[email protected] mailto:[email protected]> on behalf of Oliver Clarke mailto:[email protected]>
Sent: Monday, February 8, 2016 4:28 AM
To: [email protected]mailto:[email protected]
Subject: [phenixbb] Refine pixel size of map (for EM data)?
Hello,
I wonder whether it would be possible to add an option for phenix.real_space_refine to allow refinement of the pixel size of the map (or the unit cell dimensions - just an overall size scale factor), and write out the altered map at the end of refinement.
Although we try to calibrate this as best as we are able at the time of data collection, it is never perfect - for example, in one case I have dealt with, our nominal pixel size out of the scope is 1.19 Å, but the pixel size calibrated based on a crystal structure of a fragment of the protein is 1.25 Å. This is not a huge difference, but it is sufficient I think to have a substantial impact on refinement, particularly as regards clash assessment and H-bond/sec struc restraints.
In cases where one does not have a solved crystal structure to use for calibration, perhaps refining the pixel size in conjunction with the geometry might be of some use?
Cheers,
Oli
