Stand alone reconstruct job?

Would it be possible to add reconstruction as a stand alone job type? this would be very useful, particularly when combined with the capabilities afforded by using the low level results interface - e.g. obtaining a particle sub-selection from heterogeneous refinement, but reconstructing using alignments obtained from a consensus refinement, for example.


I have tried to “cheat” by running a local refinement but with minimal lowpass filter and minimal search range, but I get errors if I make the search range zero or too small (e.g. 0.01px, 0.01 deg gives the attached error)

Can you cheat it with a reslog job? If you set the first one to use the entire dataset…

Ah that’s a good idea! Will try

Nope - gives error

Although it does work if I set it to be one particle less than the total… bit of a hack, but works, thanks!

Actually - scratch that - it reconstructs, but it does not save the resulting volume, even in the low level results

Hi @olibclarke @DanielAsarnow - we’re adding this! I can’t think of any other hacks that could work in the current version… except that 3D var display in “cluster” or “intermediates” mode will do a straight backprojection. You might be able to set the number of intermediates to 1, and set the “width” to zero (for a top-hat weighting of particles - meaning all particles will have weight 1.0) and that might work. But I haven’t tried it!

1 Like

Has this functionality been added? Thanks.

Not yet to my knowledge


Thanks. How do you determine if your subtraction worked? With Relion, I subtract and backproject the subtracted stack. I realize I can use DanielAsarnow’s script to go back into Relion and backproject, but I thought there may be a way with cryoSPARC.

2D classification is usually a good test in that case, and has the advantage that you can separate populations of over/under subtracted particles. Even in relion I would run a quick 2D of subtracted particles - if the subtraction has worked well you should see clean 2D classes with density only for the unsubtracted region.

Thanks, the Class2D will be useful.Doing a local refine with 5deg and 1px search also works.

hi @apunjani is this still in the pipeline? It would be very useful! Particularly if it incorporated global CTF refinement params which the current Reslog job does not do. If you are going to implement, I would also suggest adding options to calculate half maps and an FSC, with optional application of a user provided mask