We recently collected the data using the graphene grid and had troubles in 2D classification.
Here are the options for the 2D classification we used. (we used the condition for low signal-to-noise particle images)
Number of 2D classes: 200
Maximun alignment res (A): 9 (similar patterns with 12, 15 Å)
Force max over poses/shifts: false (true also didn’t work)
Number of online-EM iterations: 40
batchesize per class: 400
As shown below, 2D classes look blurry without a clear boundary.
Any suggestion to improve 2D classification?
I would say there is not much evidence for clear particles on this micrograph - perhaps some sample optimization needed. How big is your particle? What defocus was this micrograph? Can you post a picture of the power spectrum/FFT?
That is what I was thinking… unfortunately, we do not have many options to optimize samples at this stage. We used -1.0 ~ -2.5 um range of defocus during the data collection.
Looking at this FFT it is reasonably high defocus, and image quality is good (you can see your graphene peaks), and ice is fairly thin based on the absence of a prominent water ring - so if particles were there, you should see them with fairly high contrast. I think sample/grid optimization might be in order. Have you tried other types of grids, or grid pre-treatments?
I agree with @rbs_sci, I can see particles but they do not seem to be picked effectively. Could you post the same mic with a 5 Å low pass? Or lower, by default the particle pickers in CS are using 20 Å.
