Unexpected results when applying rotation in volume alignment tools to volumes & particles from class3d

Hi,

When I use volume alignment tools to rotate particle sets and corresponding volumes from 3D classification (Class3D is in this case internally downsampling from 300px to 48px), it alters the orientation by a smaller amount than the input value - it still rotates around the appropriate axis (Z in the case I tested), but not by the input amount.

If I reconstruct the 3D class and then rotate the output volume and particle set in the same manner, I get the expected results.

This seems like a bug?

Cheers
Oli

EDIT: Disregard, I think this was my error

Hi @olibclarke,

This sounds unexpected. To clarify:

If I reconstruct the 3D class and then rotate the output volume and particle set in the same manner, I get the expected results.

  • In this workflow, the difference from the first paragraph is that you first use homo/hetero reconstruct to reconstruct the volume at the full box size, and then follow up with volume alignment tools on the output particles/volume?

If I understood correctly, the only difference between first and second workflows is that the first workflow has downsampled volumes fed into the volume alignment tools job, whereas second workflow has full-sized volumes; particles are the same in both cases.

To follow-up, some more questions:

  • If you take the particles from the problematic volume alignment tools job and reconstruct these: is the map’s orientation consistent with the outputs of the first, or second volume alignment tools jobs described in your post?
  • What is the amount of rotation applied?
  • Are you also shifting the volume simultaneously?

Michael

  • In this workflow, the difference from the first paragraph is that you first use homo/hetero reconstruct to reconstruct the volume at the full box size, and then follow up with volume alignment tools on the output particles/volume?

Correct.

If I understood correctly, the only difference between first and second workflows is that the first workflow has downsampled volumes fed into the volume alignment tools job, whereas second workflow has full-sized volumes; particles are the same in both cases.

Yes, that is corrrect.

  • If you take the particles from the problematic volume alignment tools job and reconstruct these: is the map’s orientation consistent with the outputs of the first, or second volume alignment tools jobs described in your post?

The first - a reconstruction has the same issue, suggesting that for some reason the wrong transform is being applied (even though it is correct in the log)

  • What is the amount of rotation applied?

90 degrees (pi/2) intended. If performed with the volume from class3d, it looks like ~70 or so (I haven’t measured exactly).

  • Are you also shifting the volume simultaneously?

No, just rotation around Z

EDIT: I think this is my screwup - rotating a lot of classes and I think I messed up assignments, going to close this for now - sorry!