Odd 2D classes while merging data

2D Classification
1

Dear All,

I am working on two datasets (same sample, same data collection scheme). I started to process dataset1 and got to a good map after refinement (70k particles at the end). I then process dataset2 till 2D classification (only 30k particles) and I wanted to merge them.
When merging them, I run first a 2D classification and get these odd classes:

J367_2d_classes_for_iteration_24
J367_2d_classes_for_iteration_241920×963 164 KB

I tried then to use the class2D_min_over_scale option during 2D classification and the classes got better:

J366_2d_classes_for_iteration_40
J366_2d_classes_for_iteration_401920×963 163 KB

But then when trying to run a refinement, or to start over from ab initio, the maps look very odd (ab initio fails, in refinement, resolution and quality map get worse).

Does anyone know where the problem could be?

Thanks a lot
Gabriella

2

Hi Gabriella

Did you try minimizing over per-particle scale in refinement as well?

Cheers
Oli

3

Hi Oli, thanks for your reply.
Yes I did, however resolution is much worse than it was previously.

Cheers
Gabriella

4

Were samples collected on the same scope, at the same mag and dose? Close together in time?

5

Yes, same scope, same mag, same dose and with two months in between (if that’s what you mean).

6

hmmm - maybe beam tilt in one and not the other? did you try Global CTF refinement?

7

No I use the same pipeline to collect the data.

No I did not try. What would be the reason to run Global CTF refinement? (sorry I am not familiar with that)

8

Global CTF refinement refines and corrects for optical aberrations which can vary from dataset to dataset (even if collected in the same way). I would try running that as a diagnostic at least.

Do both sets independently refine to high resolution, and things only get bad when they are combined?

9

I see. I have run the Global CTF ref and now using the particles for refinement.

Dataset 1 goes to 3.8A, for the dataset 2 (only 30k particles) I stopped at 2D classification and directly merged the particles. And the dataset 2 2D classification looks just as for dataset 1.

10

How did the Global CTF look (check the log file) - were there significant aberrations to correct, and did they differ between the two exposure groups?

11

Is this data you look at?

image
image933×387 59 KB

12

Yes although you want to look at the plots above it (the red and blue ones) to get an idea of whether the aberrations are being fit appropriately. If this is correct, this is a lot of tilt! And could make a significant difference to refinement at your current resolution.

Also there should be two sets of numbers in the log - one for each exposure group (which in this case should correspond to your two datasets)

13

image

image
image549×783 161 KB

image
image409×647 145 KB

I am so sorry, I am not very familiar with this yet. I appreciate your help :slight_smile:

14

Definitely there is some pretty strong beam tilt in the underlying data - correcting that should help somewhat even for the individual datasets, and may help during refinement of the combined set.

The first plot “Data for odd terms” shows if there are aberrations present in the underlying data - the red and blue indicates beam tilt in this case.

The second plot shows a model for this beam tilt, and the third plot shows the residual after applying this model. This third plot should just look like noise if beam tilt has been well corrected - in this case there is some residual at least for group 21, so estimates are not perfect yet (to be expected at this resolution).

You may want to play around with the resolution limits for the data to use in Global CTF (the defaults are quite conservative), and once you feel you have the best fit, use the resulting particle set for a fresh refinement (e.g. NU-refine).