Refinement/Classification parameters for noisy micrographs

Dear all,
We are currently processing a dataset where a relatively small complex (300 kDa) is in very noisy environment (in cell), and thus the contrast is very low - particles are overlapping, very close to each other etc. We can locate the particles but we are stuck at 6-7 A with ~1 million particles.
Are there any specific parameters for 3D classification/heterogeneous refinement/non-uniform refinement one should try that can improve the reconstruction?

Best wishes
Dominik

I’d see how much you can clean the dataset up in 2D, probably multiple times, then try using only the very best classes after multiple 2D rounds. Use a high number of classes, high target resolution and high initial particles per class, possibly with 2-4 final full iterations. Might take a few goes and playing with the parameters to find what works best. When you’ve got the cleanest particles possible, then try going back to 3D.