I have a nice actin dataset where I think I should be able to obtain better resolution, but I am stuck at 2.9A. I have 2.5M particles after extensive 2D and 3D classification (first doing a refine, then 3D classification with ~30 classes to remove any filament that is not perfectly straight). The movies were aligned within cryosparc during a live session. I did CTF refinement for defocus, then finally a helical NU refinement yields a map at 2.9A. I feel like I should go higher. What are some reasonable next steps here?
ideas:
With a non-helical sample I would expand symmetry, classify with a mask around a single sub-unit, collapse symmetry, then continue, but not sure how much sense that makes with a helical protein.
box size is currently 340A (actin diameter 70A with rise of 27A). Perhaps I could make a smaller box, since the ends of the filaments are a bit fuzzy in the recon.
reference-based motion correction. However this requires that I link the movies, but the movies were aligned during the live session. Is it possible to export movies from the live session?
higher order aberrations. I collected my data in EPU and would need the beam shift image shift values to be imported. It’s been a while since I’ve done this, is there a tutorial?
Symmetry expansion for helical data can really make the particle count explode, so I’m not convince that’s worth it…
Custom mask (use RELION) to make a helical mask focussed on the central 30-50% of the helix?
In the Details page of the Live session, there should be an “export exposures” button.
Absolutely break into beam tilt-based groups. Use the Exposure Group Utilities to split them out (you’ll need the paired .xml files if you aren’t running the latest version of EPU which (finally!) adds beam shift groups to the filename.
Depending on the data, you could see a dramatic improvement splitting by beam shift. If the acquisition was a long one, split temporally as well if chasing every last little bit of resolution.
Symmetry expansion followed by local refinement can definitely help if you have flexibility, although as @rbs_sci says, it does explode the particle count.
At the helical refinement stage, searching helical symmetry parameters over a finer grid (e.g. 512 instead of 128) can help and doesn’t take that much more time.
Refining per particle scale factors (off by default) can also help.
If using NU in helical refinement, switching off dynamic masking (setting the start resolution to 1 Ă…) can sometimes give improved results vs dynamic masking.
I’m wondering about the details of the local refinement that you used after symmetry expansion. Every time I did this, the resolution get worse compared with that generated directily from helical refinement. Should I make the mask tighter, or is there any trick that I can adopted when setting the parameter for local refinement?
I would use a pretty small search range (maybe 3deg/2Ă… or so), with recentering on, and with an initial lowpass just a bit worse than the preceding refinement (so if the preceding refinement went to 3Ă… I might start at 5Ă… for example).
Also make sure your mask is sufficiently soft, and encloses enough signal - something like the middle third or middle quarter of your fibril might be a good starting point.
Here are the parameters that I usually set for the local refinement job.
Generally, the mask was created in Chimera, with 3/4 of the fiber being covered. And then it was processed in cryoSPARC with Dilation set at 2 and Soft padding set at 12.
Right - so these search params are likely way too big for a helical dataset. You want to make sure the search range is smaller than the rotation/translation required to superimpose 1 asymmetric unit onto the adjacent one.
I would also probably mask more like 1/4-1/3 of the fiber rather than 3/4 - the whole point of the local refinement after symmetry expansion is to take care of flexibility of the fiber.
Also, you are using C3 symmetry - but I assume you already expanded in C3 + helical symmetry? In which case you should use C1 in local refinement
