When using the “Downsample particles” job with the “Recenter using aligned shifts” setting enabled, the resultant .cs file shows that the alignment shifts are reset to sub-pixel values (presumably to match the new images), but the particle coordinates/locations are not updated.
Thus if these are later re-extracted, or the original particle.blobs are attached, they reconstruct with the original centre but with loss of aligned shifts, resulting in a poorly aligned average. This is particularly apparent in the case of large alignment shifts, such as a recentering by volume alignment tools, which is also the main case for enabling recentring during downsample.
I’m not sure the issue of being able to revert recentred, downsampled particles using the original particle.blobs is easily surmountable since the alignment shifts do need to be reset to align the new recentred & downsampled images properly, but certainly upon re-extraction I think the expected behaviour would be to extract boxes centred around the new centre, rather than the pre-downsampling centre, so the particle locations would need to have been updated.
I suppose a workaround in the mean time is to first re-extract with recentring and then downsample without afterwards.
Thanks for reporting this – you’re right that location metadata are not currently updated by Downsample Particles when re-centering is enabled. Re-centering was initially added to Downsample Particles to address the case when one wants to recenter particles on a new arbitrary center but either:
doesn’t have access to the micrographs (e.g. particle import from EMPIAR, hence can’t use extract from micrographs, and locations don’t exist)
wants to recenter particles that have been subtracted already, hence can’t go back to micrographs
In most other cases it may be possible (and recommended) to re-center, crop, and/or Fourier-crop all within the “Extract from Micrographs” job. In case 2 it is worth updating the locations, so I have recorded this. Thanks again for reporting and definitely let us know if there are other cases where this poses a problem
I want to recenter and crop a set of particles-images but I am having problems getting the expected results.
I used the downsample job with the option to recenter particles according to previous alignment shifts. I was then expecting than that doing an homogeneus reconstruction only with the recenterd particles as input would esult would result in a recenterd 3D map, but this is not the case. Please advise how to use this tool (or others) to do the expected task of recenter all particles according to a previous alignment and crop them around the new center.
How did you recenter prior to downsample? I mean how did you get the appropriate shifts? I would use volume alignment tools to recenter, then use the particles from that for Downsample as you describe. Then it should work as you expect.
When doing the volume reentering in the volume alignment tools would the output include the reentered particles (not masked) ? Do I have to turn on reasign UIDs ?
The particle images output by volume alignment tools are the same as the input, but they are bundled with the shifts that allow you to extract around the new center (using Downsample or Extract from Micrographs). Masks are not relevant to this question - masks are (typically) not applied to the particles themselves, but to the reconstruction. No, you should not need to reassign UIDs.