No protein features seen in 2D classes but rather some weird rings

2D Classification
1

Dear cryoSPRAC users:

I’m a beginner to learn and use cryoSPARC version 3.1.0. Recently we collected a dataset on Krios microscope with K3 detector, the data acquisition settings were:
Total dose: 80 e/A^2
Dose per frame : 2.0 e per frame
40 frames/movie
Defocus range: -0.5 to -2.25 um (interval 0.25 um)
Magnification: 130 kx
Pixel Size (Sampling): 0.675A
Exposure time: 2.3 secs
Spherical abberation (Cs): 2.7
Voltage: 300kv

The protein particle is ~170 kDa and has a size of ~60-160 A in diameter, appearing a bent-rod like shape. I chose elliptical blob for the blob picking on a small batch (~10%) of micrographs and used the following parameters:
Minimum particle diameter: 70 A
Maximum particle diameter: 170 A
Turn off Use circular blob
Turn on Use elliptical blob
Other settings were kept as default.

After inspecting the picks, I did particle extraction using the parameters:
Extraction box size (pix): 400
Fourier crop to box size (pix): 320
Other settings were kept as default.

For 2D classification, the parameters were set as:
Number of 2D classes: 200
Number of final full iterations: 2
Number of online-EM iterations: 40
Batchsize per class: 400
Other settings were kept as default.

But the output 2D shapes don’t appear any protein feature but rather some weird rings, as shown in the attached image.

P4_J14_2d_classes_for_iteration_41
P4_J14_2d_classes_for_iteration_412397×4800 2.07 MB

Could anyone please give me some clue? I couldn’t figure out which step/setting I did wrong.

Thanks!

2

Have you got any answer?

3

While I don’t think this will improve the classes significantly, did you try turning OFF “Force Max over poses/shifts”?

Also, could you please share a raw micrograph as well as one where we can see the picks?

4

agree picks probably still wrong from blob picker. use class 5, 7, and 32 to run template picking. adjust power and NCC thresholds to keep ~half as many particles as are picked, such that it keeps those which most closely match the templates. then run 2D again and consider new ways to proceed. Fourier cropping 400 down to 320 seems ineffectual, I wouldn’t do this, either cut it to ~100 (2.6Å pix) to save processing speed, or keep it full resolution. A slightly smaller box will also help in early steps to drive alignment and ignore neighboring particles - consider 320 as extraction box size or even 256.