Seeking suggestion regarding particle picking

Particle Picking
1

I collected a dataset of ~15,000 micrographs on a Titan Krios at 300 kV (higher magnification / smaller pixel size). The target particle size is ~12–14 nm. After Patch Motion Correction and Patch CTF Estimation, I tried blob picking on ~4,000 micrographs, inspected the picks, and ran 2D classification, but I didn’t get any meaningful classes.

Next, I used a previously resolved map of what should be the same complex to do template-based picking. That yielded only a small number of plausible particles. I used those to train Topaz Train, but Topaz pick/extraction is very sensitive to the expected number of particles per micrograph parameter changing that value produces very different results including picking Junk particles. Visually, I can see a lot of particles in the micrographs, but I can’t pick them efficiently.

What would you recommend trying next in CryoSPARC? For example, should I:

  • curate more positive/negative examples and retrain Topaz,

  • adjust the Topaz threshold/expected-particles settings,

  • downsample/rebin to better match the 12–14 nm size, or

  • revisit CTF/defocus ranges and picking radii/diameters?

In fact, I have done all the steps as mentioned above. For some reason, I am not getting the particles I want to pick. pixel is 0.59 and mag is 165 K

2

Try denoising, see if it improves things.

Manually pick 20-50 denoised mics and 2D class that (low class count), see if it looks like anything more than noise.

A representative micrograph, unlabelled and labelled with picks, and some 2D classes would assist with potential diagnosis.

3

@rbs_sci thanks so much! The number of frames were 17 and number of fractions were 83. I am not sure if they are causing issues with the motion correction and thereby the downstream process. I have followed your instruction but actually could not figure out it properly.

J229_denoised_output_micrograph
J229_denoised_output_micrograph810×810 305 KB

J229_denoised_output_micrograph (1)
J229_denoised_output_micrograph (1)810×810 308 KB

J224_rigid_motion_for_008295344470181586649_foilhole_27821762_data_27809601_0_20251028_143205_fractions
J224_rigid_motion_for_008295344470181586649_foilhole_27821762_data_27809601_0_20251028_143205_fractions983×358 54.6 KB

J224_rigid_motion_for_002285983854477562395_foilhole_27821763_data_27809544_1_20251028_143210_fractions
J224_rigid_motion_for_002285983854477562395_foilhole_27821763_data_27809544_1_20251028_143210_fractions983×358 46.2 KB

4

The background looks rather strong for denoised mics - is that imaged from ice in the holes? What were the buffer conditions? The 2D classes would also be useful. :slight_smile:

6

20 mM Tris, 100 mM NaCl, 5 mM DTT, 10 mM MgCl2, and 8 mM CHAPSO ( to avoid preferred orientation bias). We got 2.46 A before with the same complex but with different DNA (WT type). This time only pixel size and mag was changed. We target edge for data collection as detergent brings them down there XD.

J458_2d_classes_for_iteration_120.png
J458_2d_classes_for_iteration_120.png1920×1537 280 KB

7

Figured out the problem. I had to do several rounds of 2D classification, which resulted about 50 K particles. Resolution is now 3.96 A. Will recollect data again.