Not getting useful 2D classes


I have a dataset from a supposed dimer of 100 kDa. I expect the length of the protein to be about 120 A on the longest side. I’ve been trying to get 2D classes but it didn’t work well. First, I was a bit strict on the blob picker and tried to narrow down the picking to reasonable particles and the 2D classes seemed to be overfitted noise:

Then I tried 2D classification using much more particles and I did the elliptical blob picking. It didn’t help much.

The 2D classification jobs parameters are default except:
Force max: off
Online EM-iterations: 40
Batchsize: 500

I’d appreciate if anyone can give me some ideas how to approach this. Thanks!

How do the micrographs look? Can you clearly see particles in them?

1 Like

take the 5 best (albeit overfitted noise) shapes and use them to template pick. for something that small, blob picking will pick every visible dot in the micrograph. I would focus on getting good picks. For a test case, use the filters/thresholds to take ~40% of total picks and run the 2D with 500 batchsize. If you get a few good templates, repeat the process with new good templates.