I followed topaz train, topaz extract, inspect, extrat from micrograph, 2D class, selected 2D class, Ab initio, homogeneous refinement, and NU_refine, and obtained 3.37A, with C13 symmetry.
Two issues I am facing
As the protein is 13mer donut-shaped, I am not getting a complete structure with C1, while C13 produces a clear image but with a tiny ring/dot-shaped structures passing through the donut structure in rod shape. (trying heterogenous refinement not showing such small dot/rings but resolution is 5.84A)
I can see 5.84A Cryomap and protein structural model fit in chimera using volume data, fit in map, but when I am using this for real space refinement in phenix it shows map and model not aligned, autosharpning in phenix showing kurtosis 61.89 with surface map area 0.39.
Should I focus on more refinement with C13, 3.37A with artifacts, or can I do something with the 5.84A map without artifacts
so it’s a 13mer, but doesn’t exactly obey C13 symmetry? I would suggest refinement in C13 followed by symmetry expansion, then local refinement in C1 using a mask around one segment of the “donut”. When you say the C13 has a clear image, do you mean you can see sidechains etc?
Thanks for you quick response. Yes, I can see 13 side chains arranged in a whirl, and look similar to model protein structure. Running with C13 gives complete structure with equal density but with such artifacts I mentioned above. I am not sure if I am on right path. I also run local refinement which slightly improved resolution as I can see in GSFSC plot, but not sure how to make sure C1 in local refinement as could not see option.
Right - I am suggesting after C13, run symmetry expansion. Then, use the sym expanded particle set and C13 map to initialize local refinement, using a soft mask around just one segment of the donut. Local refinement by default will be in C1.
i will follow your suggestion and update here, below is a snapshot of issue marker with arrow, multiple refinement reduce this dust around structure, but this marked tiny structure possibly making further refinement, alignment with map or rigid body fitting difficult
This looks like it might also have other issues - looking at the density I suspect preferred orientation. What does your orientation distribution and cFSC plot look like? Do you see a diverse range of 2D classes, or one dominant projection?
from tilted images, i can see images from different orientation, mostly from front, some from sides, but not from back/bottom side, and front view prominent (but tilted).
Yes, this looks like you have a significant preferred orientation. In the case of a particle like this, the top views (donuts) may be easier to pick, so it may be worth exploring training a Topaz model to selectively pick side views; otherwise you may need to modify sample prep (high CMC surfactant additives) or data collection (tilt) conditions.
I understand, although I manually pick nearly 1500-2000 particle, and based on this trained topaz, followed by topaz extract, I tried to select multiple orientation as possible, but might be only top view picked and trained, I think i need to work on manual picking? what if side view are blur, picking clear top and blur side view would be helpful to improve?
Sounds like you are on the right track, but yes, I would take the particles from your side view 2D classes, inspect and curate them on the micrographs, then use only genuine side view picks to train a new Topaz model to preferentially pick side views. You can then always merge the results (hopefully with more side views) with the top views you already have.
Great start @KMS. @olibclarke is 100% correct, symmetry expansion, local refine in C1 is the standard route, but as he mentions visual inspection of the map and the 3D-FSC/low cFAR suggest anisotropy. The GS-FSC suggest there is more resolution to squeeze out: use full box size or increase box size ?
I found in a similar system with extensive 2D classification, select 2D (only take the side views), and retraining a TOPAZ model with just the side views helped. Also, rebalancing orientation, gives particles that are more balanced for TOPAZ training.
In these ring-like systems the top views do no help that much, it is more of the side views that yield useful reconstructions. The 20S proteasome is a good example of this.
High tilt (30-45 deg) could help if you don’t want super high resolution and can find an acceptable dose. Sub-tomogram averaging is not widely for these systems but promising.
