I got good resolution structures from a good dataset.
But I have a plan to change a pixel size of the same dataset.
You know If I re-run from the beginning as ‘import movies’, it takes very very long time.
So, I thought that particle (x,y) information of the good resolution structures can be used after changing a pixel size.
Is it possible ? How…? particle (x,y) information where is it ?
1 Like
Do you mean the pixel size is wrong? If it is a minor error, it can be fixed in post-processing. If the defined pixel size is dramatically different, however, it will affect everything - from dose weighting in motion correction, but most importantly the CTF estimation. If you dramatically change the pixel size the CTF information can no longer be trusted, which is significantly more important and will have a serious effect on everything else.
1 Like
I run processing; pixel size was 0.87.
The real pixel size is 0.81. Do you think that is highly different between them ?
That’s nearly 7.5%, so yes, it’s pretty different.
It won’t have such a great effect at lower resolutions (depending on the correlation between the experimental and simulated CTF at different Thon rings) but will make a big difference at higher resolutions. It can manifest in different ways; e.g.: a reconstruction which reports high resolution but doesn’t “look” high resolution or odd dips in the FSC curve, which can often be fixed by beam tilt refinement but it is better to solve by correcting the cause. Astigmatism makes this worse. Danev et al (2021) show the effect on FSC for apoferritin (Fig. 2).
If there are any PDB models, determining how far off the pixel calibration is, is pretty easy, but if it’s a complete unknown it can be a bit tougher. Fitting a single alpha-helix can be enough to correct if it runs for long enough. It sounds like you already know how far the calibration is off, though.
If you plan on doing Bayesian polishing in RELION, that will correct for the variation in the dose, but I’d think about redoing CTF estimation, 2D classification and 3D refinement with corrected pixel sampling. If particle picks look good, just re-extract using the micrographs with corrected CTF parameters. I think you’ll need to re-import the motion-corrected micrographs with the corrected pixel size (Import Micrographs, point it at the folder where the motion corrected micrographs are, set pixel size etc., enable “Skip header check”, I think is the quickest way)… I can’t see a way to force pixel size in the CTF jobs.
The results may change significantly. You may well find a big jump in resolution and/or map quality.
2 Likes
Thank you for your post.
Your post has a great detail and will be very helpful for my job.
@rbs_sci fyi if you can refine the Cs, that will account for pixel size error in CTF evaluation. I like to compute the real pixel size using a high-res apoF dataset, I think that way is better than manually testing different sizes and correlating to an x-ray structure.
https://relion.readthedocs.io/en/release-3.1/Reference/PixelSizeIssues.html
@shinjw1887 rbs_sci is right, 7.5% is a lot. If your particle would be 100 Å across there would be a 7.5Å error on the edge vs. a correct reference! With that much error I would re-run everything, since the initial dose-weighting is off by even more, nearly 15%. (Dose is per area).
1 Like
Thank you for your suggestion. I’m doing re-processing as you said.