Hi all,
I am wondering when to use “Window dataset” for 2D and 3D. I understand the principle: circular masking the cropped particles to improve cross correlation. For 2D this is distinct from ‘circular mask’ which applies a mask to the 2D class averages.
However, I have tried altering the “Window Dataset” but it doesn’t improve the map at all, even when the particles came from highly packed micrographs. It’s a bit confusing, since I would expect that correctly using this parameter should improve alignment since it occludes neighboring particles, but it seems to make no difference. Is there an explanation for this, and can someone explain when they think that modifying this parameter might actually have a noticeable impact?
Thanks!
You don’t want to use window dataset (which windows the particles), you want to change the circular mask radius (which is applied to the classes) - try that? Definitely helps with crowded particles!
Hi Oli, many thanks. I understand that the circular mask is applied to the class averages, I guess my question is more conceptual: when should the “window dataset” option be used?
There are two places where a circular mask can be applied: to the 2D class averages (“circular mask radius”) and to the raw cropped particles themselves (“window dataset”). Since the alignment algorithms perform cross-correlation between the 2D class averages and the cropped particles, why isn’t it common practice to adjust the “window dataset” parameter so that its radius roughly matches the “circular mask radius”? That way, both the particles and the class averages would have the same circular masking applied.
In practice, though, it seems that only masking the class averages has a noticeable effect. I’m curious why that is, and under what circumstances the devs recommend using “window dataset.”
Perhaps more importantly, in 3D jobs there is no “circular mask radius” because there are no 2D class averages to mask. In that case, it seems like applying a circular mask directly to the particles via “window dataset” would be even more beneficial. Is that the intended use case?
There is a spherical radius in some 3D jobs (e.g. ab initio & heterogeneous refinement).
Windowing the particles happens to the raw particles, meaning before CTF correction - so if you aggressively window, you will cut out delocalized information (particularly high res info). By contrast, because the circular/spherical radius is applied to the class averages/volumes, this is not an issue in that case.
Cheers
Oli
1 Like
thanks, didn’t realize windowing happens BEFORE CTF correction. Seems like a bad idea. Wonder why it’s done that way…