Hello everyone,
I have a few negative-stained micrographs with a pixel size of 4.65 Å/px. I generated 2D templates via “create templates” job from a 3D map (collected on Krios) that has a pixel size of 0.826 Å/px, and then used those 2D templates for template-based particle picking. After template picking and running 2D classification, the 2D classes look extremely poor (almost no recognizable particles, mostly noise).
Does the pixel size mismatch between the micrographs (4.65 Å/px) and the 2D templates (0.826 Å/px) cause this problem? If so, does CryoSPARC handle pixel size for template creation and template picking (does CS automatically rescale, or do I need to match pixel sizes myself)? What is the correct way to match pixel sizes so that 2D templates generated from a high-resolution map can be used for template picking on lower-resolution negative-stain data?
Hi Abhipsa, when you imported the micrographs, did you switch on the “negative stain data” flag?
And when you inspect the micrographs, can you see clearly recognizable particles (and the template picks hitting them)?
Scaling of the templates shouldn’t be an issue, this is handled automatically, although I would recommend blob picker over template picker for negative stain. The reason for this is that because resolution is inherently limited in NS, any template bias will be harder to detect (as there is no higher resolution signal present to allow discriminating real from Einstein-from-noise classes).
Yes, when I imported the micrographs I toggled on both “negative stain data” and “output constant CTF.”
The micrographs themselves look surprisingly good. I can clearly see well-defined particles.
I tried three different approaches:
Manually picking ~1200 particles and then using those for template picking.
Blob picking followed by template picking.
Template picking using 2D templates generated from a 3D map.
The best 2D classes I have obtained so far are from approach #1.
(I had actually thought blob picking was not ideal for negative stain, since there are no aligned images and I assumed it would be hard to get a meaningful Gaussian signal).
However, in all three approaches, though, I am seeing the same issue at the template-picking stage: when I inspect the picks and adjust the NCC and power thresholds, the picks mostly land on the background rather than on the visible particles. That’s what made me start experimenting with blob picking and the 3D-map-based templates, but the behavior looks very similar in each case.