I downsampled by particles to a box size of 256 for local refinement using the “downsample particles job” because I was having memory issues. I want to get the particles back to full resolution- how exactly would I go about that?
I think you can re-extract your particles using “Extract from Micrographs” at a box size you want without Fourier cropping option. Your full resolution particles should contain 3D alignment information from local refinement even after re-extraction.
Awesome, I wasn’t sure if it was going to keep that alignment information so this is good to know!
You shouldn’t need to re-extract - you can just replace the particles blob with one of the outputs from the original extraction job using the low level results interface (see elsewhere on the discussion forum)
@olibclarke I tried to do that, and then the program crashed. I saw the previous discussion post, and I was a bit confused on how exactly to do that to be honest, lol
Hi @merandamasse, here is a screen recording:
Here I am going in the other direction (replacing original particles with downsampled) but it is the same idea - does this make it clearer?
Basically if your job is in the “building” state you can expand the particles component to reveal subcomponents (corresponding to particle images, alignments, CTF info, etc). Those subcomponents can then be individually replaced with subcomponents from another job (in this case the outputs tab of your original extraction job).
Yes this makes so much more sense! Thank you so much, you rock!
No problem! this is also very useful for a lot of other things - e.g switching back to do a reconstruction using global alignments after local classification, replacing subtracted with non-subtracted particles, or transferring higher order CTF params between particle sets.
quick question: Extract unbin, refine, then bin extract with re-center shifts, classify, now refine best class unbin by using original extraction job. To do this I replace low-level particle.blob. Does this replace also back to the original extraction position of the particle (I hope not)? Or is that data in particle.location input so it is still the corrected “pick”?
The particle blob only refers to the particle images as far as I am aware, with the location information in a separate blob.
Related to this thread, in Cryosparc, if I used ‘Fourier crop’ in the original extraction job to downsample the particles and used it for 2D classification, is there a way to link the full size particles to subsequent jobs without re-extracting? Thanks for any comments.
No, because these full size images don’t exist. You have to re-extract the particles without downsampling. This also gives you the opportunity to re-center the good particles based on their aligned shifts, either from the 2D classification or any 3D refinement you’ve done. I prefer to run a 3D refinement (it’ll be fast since box is smaller) to make the aligned shifts are all consistent and accurate.
Seconding @DanielAsarnow’s reply – assuming you have used the
Fourier-crop to box size parameter to set the bin size. The full size particles will not exist if this parameter was set, and can only be generated via re-extraction at the full size. The link/correspondence between these particles and the F-cropped particles always exists, because the particles share the same UIDs – the only difference is that they have different
blob output fields.
If you instead set the box size, left the first F-crop param empty, and then set the
Second (small) F-crop box size parameter to the smaller size, then both the full-size and the downsampled particles will both be output from the extraction job. You can use the
blob from either result interchangeably via the low-level results feature of the job builder.
Greeting there Oli,
I have concern here about replacing the downsampled partilces with full size particles:
Let we say I have done Ab Initio and hetergeneous refinement using downsampled particles and choose one class for further homogeneous refinement. I will re-extract my particles of this class with full size and has re-center option opened. I don’t think I can replace only the particle blob when I build the homogeneous refinement job, because the aligned shift from hetergeneous refinement already applied during re-center, and the alignments 3D information file from hetergeneous refinement is still kept in the low lever interface, which will cause wrong starting if I was right.
Sorry if I was wrong, I’m from cryoET filed and green to single particle.
Thanks and cheers,
Hi Yanhe - I’m not sure I follow… can you explain a little more?
The orientations from hetero don’t matter in this instance because they will be scrubbed upon starting homogeneous refinement in any case - I’m not sure I’m understanding your question though?
Thank you so much in advance. I understand that the orientations will be scrubbed. But how about shift information from hetero? My understanding is alignments3D file include both orientation and shift information for particles, right? And the re-center is calculated based on location file and shift information in alignments file.
Thanks and cheers,
alignments3d altogether (shifts & rotations) will be reset during homogeneous refinement - but I’m still not entirely sure I’m understanding your question completely
Thanks for this informative reply, it make more sense to me if the alignment3d altgether (shifts and rotations) will be reset.
Thanks and cheers,
And I have the reply by DanielAsarnow to huqi from up pasted as below, so it means that if I didn’t do the re-extraction with re-center option, the accumulated accurate aligned shift information by 2D classification or 3D refinement will be lost, since the shifts will be also reset, am I right?
: No, because these full size images don’t exist. You have to re-extract the particles without downsampling. This also gives you the opportunity to re-center the good particles based on their aligned shifts, either from the 2D classification or any 3D refinement you’ve done. I prefer to run a 3D refinement (it’ll be fast since box is smaller) to make the aligned shifts are all consistent and accurate.
Thanks and cheers,
Dan’s reply is referring to the fact that if you initially extract with Fourier cropping, you cannot link back to the full scale particles, as they don’t exist (they were never extracted in the first place). I am not sure how this is related to re-extraction with recentering in your query?