Extremely light or dark levels in 2D-class results of the particles from two datasets?

2D Classification
1

Hi, everyone,

I collected two datasets with same pixsize and merged them together for processing. Unfortunately, we observed some weird 2D classes with extremely light or dark levels, however, the light or dark classes look like my target particles. Can someone meet this problem? Something wrong about grayscale? How to solve it?

Thanks!

Niu

2

QQ图片20230223161947

3

Please can you describe your upstream workflow, including details on how you “merged” the datasets.

4

I could venture a guess, hopefully Wolfram can get you sorted: accept them all, move on, scale factors will be automatically adjusted downstream, then you can always rerun 2D if you want to use it as a selection criteria.

5

The two datasets were collected in one same microscope with same settings (including pixsize, frame and total dose) while different lacey grids and different times.

We seperately motion-corrected the movies in each dataset using Patch-motion correction (B-factor:150) and then merged the micrographs in two datasets together for CTF-estimation using CTFFIND4 with default settings. After removing the junk images, we picked out the particles by using Blob picking, template-base picking and topaz-trained picking in sequence, and finally did 2D classfication with the extracted particles. Here we obeserved the weird classes with extremely light or dark. In addition, it seems that the light classes are from one dataset while the dark from the other one.

We also tried to separately process each dataset. It can work well for each one, but still have the same problem if we merged the two for 2D classfication.

6

We will give it a try and see what happens. Thanks!

7

Hi @niu ,

I agree with CryoEM2, this could be “normal” in the sense that the source micrographs could have a wide range of brightness levels. If you accept all good-looking classes in which your particle is recognizable, this can be dealt with downstream. You may want to switch on the “Optimize over per-particle scale” parameter in your downstream homogeneous refinement (it is off by default). There is a similar parameter for 2D classification itself which you can definitely try, but in general this works much less reliably than it does downstream in 3D.

–Harris