Possible bug: Particle pre-processing window affects corrected FSC?

I am noticing that the particle pre-processing window affects the corrected FSC. Is this the expected behavior, or is this a bug.

The FSC curves below show two non-uniform refinements that are identical, except that in one the windowing was left default (0.85-0.99), and in the other windowing was turned off. In the windowed one, the characteristic drop in the corrected FSC, where the phase randomization begins, is much more dramatic. Other tests show that the smaller I make the window the worse this gets (despite the fact that the window is still larger than the object I’m refining). This occurs both in non-uniform and in homogeneous refinements.

Bug, normal, or am I doing something wrong?

Running version 2.15.0

Windowed 0.85-0.99:
P67_J165_fsc_iteration_008 (1)

Not windowed:

As far as I know, this is not a bug. I’ve noticed this too and have tested it with various values. In my case, the decreasing the values to about 60% (0.6-0.65) actually worked wonders. It improved the shape of the FSC and the overall features of the map. I’m attaching some examples of what tests I ran and although I did not use the values that you see here finally, I do clearly see the effect. Also, having a tight cutoff (for ex: 0.4-0.41) is generally speaking detrimental.


1 Like

It’s not clear to me why the preprocessing window should even affect the corrected FSC at all. My only hypothesis is that the particle picks are not well centered (which is definitely possible), so that the particles are getting cut off by the preprocessing window. If this is right, then the changes to the FSC are essentially a masking artifact.

If that’s true, it points out the need for some mechanism for recentering the particles on refined coordinates rather than the original picked coordinates, which cryoSPARC can’t do, as far as I know.


Hi @rj.edwards,

Regarding the effect on FSC, using a very tight window could induce high-frequency artefacts in the particle images, which will interfere with the high-frequency information that is used for alignment. I can’t comment on the other effects that have been mentioned, but perhaps @apunjani can provide more insight.

As well, re-centering based on refined coordinates will be available in the next release of cryoSPARC, which we hope to release very soon. It should also be noted that this release will also feature a reconstruction-only job, which will allow you to reconstruct the same aligned particle stack with varying window parameters. This could help see if the windowing is negatively impacting just the alignments, or just the reconstruction.


1 Like