I have two questions on processing helical data on an amyloid sample. Both ask essentially the same question about how to calculate a helically-symmetrized map without refinement, but they also have other subtle flavors. My attempts at searching did not yield any answers.
After Global and/or Local CTF refinement, I would like to simply calculate a new model using the new CTF values but without performing any further refinement. I don’t see how to do that. What I find is that the CTF refinements followed by more helical refinement result in worse density with poorer separation between layers. I want to try individual CTF refinements, especially anisotropic magnification and defocus, to see any effects, but the backsliding after subsequent refinement may be masking them. Or the CTF refinements are going bad.
The Relion advice is to pick long for achieving a decent model where the longer filaments align better, then re-pick shorter segments for improving resolution by avoiding bendiness within particles. Again, after re-extraction, and assuming that the new shorter particles are still appropriately aligned (are they?), I just want to re-calculate a new map without refinement. As for (1), refinement backslides to a poorer map with merged density.
Perhaps there is an obvious answer, but any pointers appreciated.
Hi @jxc100, thanks for the post. To answer your questions:
You can accomplish this via the “Homogeneous Reconstruction Only” job. It simply takes the particles with their current ctf and alignment estimates, and constructs the map without refining either ctf or alignments. You can optionally specify the symmetry as a parameter to this job.
You can re-extract particles at a smaller box size (giving you shorter segments) by passing them through the “Extract from micrographs” job; by default, they will be re-centered and then re-extracted. You should then be able to run the Homogeneous Reconstruction Only job as indicated above. Regardless of whether re-centering is enabled or not (this is a parameter of the extract job), the alignments should be preserved in accordance with the re-centering, such that the calculated map is unaffected.
If you have the “Limit shifts along the helical axis” parameter of helical refinement set to True, then re-centering is viable. I would advise though that if this parameter is disabled, its best to avoid re-centering during re-extraction to guard against the case where two particles have refined such that their alignments make them nearly coincident.
Also, I believe the “Limit shifts along the helical axis” parameter controls the “backsliding” that you are referencing. It certainly may be worth it to experiment with turning off this parameter if you think it may be harming resolution. Typically this parameter improves resolution only in the case where you have precise knowledge of the symmetry parameters, because it relies on the symmetry parameters to restrict particles to one asymmetric unit.
Please let me know if this answers your questions.
Thanks, yes, very helpful. I can now repeat the final reconstruction of an earlier refinement using the same refined helical parameters. The result is slightly different, which I assume is due to use of the previous final mask, whereas the previous job may have used some intermediate mask that is essentially the same but not identical.
However, I can’t say the same for the shorter extracted segments. The original filament is skinny and in a 512-pixel box. I re-extracted into a 256-pixel box (in both cases no binning) now that I have a good model and to avoid bendiness. As you advised, I had already disabled “Limit shifts along the helical axis” in the previous long-filament refinement, and I had disabled “re-centering” in the re-extraction. In the previous refinement, I had “Limit shifts along the helical axis” enabled as we believe we know the helical symmetry (I could test it) but I don’t know if this would affect reconstruction-only of the newly-extracted short filaments. I’m not sure why the short-filament map is poor compared to the parent long-filament map.
There is another parameter that I don’t fully understand - Helical symmetry order - in “Homogeneous Reconstruction Only”. Previously a student set it in refinements to 25 (which seems large), and varying it to 30 or 20 reduces map quality, while varying it form 22-28 has little effect. If my rise is 14.8 A, and the map dimension is 425 Å, then there are ~28.7 subunit rises over the length of the boxed filament, but I am not clear if this is directly connected to that parameter. Looking back, it seems that Filament Tracer had the Filament diameter set to 90 A and the “Separation distance between segments” at 0.1 diameters, or 9A. The tooltip for Helical symmetry order suggests 9A/14.8 → 0.6, which is hugely different from 25. If I use None (since Filament Tracer was used) the resulting map is again poor.
Back to my original purpose - I have tried the CTF refinements followed by reconstruction-only and the global CTF parameters (tilt, trefoil, magnification) do little, but the local CTF (defocus) makes the map significantly worse. I did not expect this, but at least its obvious where to focus attention.
After re-extracting at a smaller size, did you re-refine the particles, or go straight into a homogeneous reconstruction only? Sometimes helical refinement performs better with larger boxes simply due to the additional signal in the images, which helps alignment. In any case, would you be able to DM me FSC and slice plots of the final reconstruction for the parent bigger-box filaments and the smaller-box filaments that had a poorer resolution? I am also curious about the resolution discrepancy.
The helical symmetry order tells CryoSPARC how many times each image should be used during reconstruction. It should be equal to the number of new asymmetric units held in each image. This is described in more detail in the Helical Refinement guide under “Maximum symmetry order to apply during reconstruction”. Generally its recommended to pick particles such that the distance between adjacent picks is some known multiple of the helical rise. E.g. if i had a rise of 14.8 Å, then perhaps I would want adjacent picks to be 3*14.8 = 44.4 Å apart, meaning the helical symmetry order is 3. Then I would scale the “Separation distance between segments” parameter to be such that the separation distance is 44.4 Å — so divide this by the diameter of 90 Å to get a separation parameter value of 0.49. I chose the order of 3 arbitrarily, but you can experiment with greater values.
PS: what is the resolution of the structure reaching? Amongst all refinement algorithms, CTF refinement relies the most on high-res signal, so is most likely to fail when the data is still at moderate or low res.