First of all, my protein is about 12nm in size.
In my opinion, the particle state in the micrograph looks fine, and the results of patch motion correction or patch CTF don’t look bad.
However, when I proceed with 2D classification, I get the result like the picture attached below.
I am very sad that I can’t proceed with the next process. Is there any way to solve this?
I first tried to make a template using the already known structure and do particle picking, and I also tried to make a template using the manual picker, but the result was the same.
Micrograph is attached. (total collected micrographs: 1,163 micrographs)
How many particles do you have in total? Probably 500,000 - 1,000,000 or so with 1100 micrographs and the number of particles visible?
First thing to do is check that the picks look fairly well centred on your particles. If they are, then in the inspect picks job, adjust the lower boundary on the slider to avoid any “blank” picks where you can’t see a particle under the circle.
It looks a little heterogeneous.
Did you use the default parameters for 2D classification?
If so, try the following:
Number of 2D classes: 200
Initial classification uncertainty factor: 5
Circular mask diameter (A): [slightly larger than your particle; 130 or 140 Å?]
Number of final iterations: 3
Number of online-EM interations: 30
Batchsize per class: [vary based on total particles, try to aim to cover the whole dataset about 1.5-3 times in the small batch iterations, I often use 150 or 200]
Perhaps also try another run with the same changes, with two other things enabled:
Enforce non-negativity: on
Use clamp-solvent to solve 2D classes: on
The above will slow classification down a lot, though.
What parameters are you using for particle extraction? Also what are the other expected dimensions of the particle other than 120A? What is the pixel size of the micrographs you are using?
Hi @chuckchuck! I understand your frustration, your micrographs do have clear particles but those 2D classes aren’t looking great. Let’s see if we can improve the results for you!
I think my suggestions will be similar to what @rbs_sci suggested above — have you tried those parameters, especially increasing the number of classes?
I also wonder if could post a screenshot of the micrographs with the particle picks indicated — I’m wondering if there is a lot of “junk” that is contaminating your good picks. The easiest place to find such an image would be near the top of Extract from Micrographs or Inspect Particle Picks jobs.
It looks like non-negativity and clamp solvent helped a lot!
It looks to me like a somewhat arbitrary part of the edge of your particle is being chosen as the pick, which moves the rest out to the edge of the mask. Have you tried using a ring-shaped blob instead of the default (i.e., turning off “Use circular blob” and turning on “Use ring blob” in the Blob Picker tool)? This may help center your particle picks.
It also looks, to me, like the box might be a little small for the size of the particle. What size is your box (in angstroms)?
And finally, have you tried going straight to ab initio with your particle set? It looks to me like a decent proportion of your particles are correct, they are just off-center picks. Ab initio might be better able to filter and center the picks, and from that you could make templates of the whole target to use in Template Picker.
I’d select a few classes which are nicely centred and re-pick with those templates. They don’t have to be particularly high resolution classes (they’ll get filtered down to 20A blob territory anyway). Projecting templates from a good-looking ab initio model is a good strategy. But if using a particle set that heterogeneous for ab initio I’d think about asking it to generate 15-20 classes… which will probably take a while.
@rposert ,
As you suggested, I’ll try to progress up to the 2D class again, using a ring-shaped blob instead of the default values.
And the size of the box is 180A. In the Extract micrograph, I set it to 256px, which is the default. Next time, I’ll set it to 360px, which is a larger size.
I also tried using Ab initio with symmetry set to C1, but the map shape looked too weird, so I didn’t make it a template.
@kstachowski
Oh, I thought I needed a high resolution class, but I guess I was wrong, I’ll take that as a note and try again.
Using high resolution references for template picking risks template bias and picking noise. While current 2D classification methods should help alleviate any “non-particle” picks, it’s still a risk (along with wasting storage space and processing time!)
What you want is a class that looks like “something” - that you can recognise as something other than just junk (be it ice, aggregate, noise blob, and soforth - but it’s good practice to low pass filter (both the templates and the micrographs, although if one is done the other is effectively a simple safety measure/sanity check).
