I am currently processing cryo-EM data for a protein–DNA complex structure using cryosparc v4.7.1.
I have encountered an issue where the 3D Variability job takes an unusually long time to complete, it took 2 days to complete 7 out of 20 iterations.I would like to ask for your advice on how to optimize its runtime.Specifically, could you please suggest settings for parameters such as: number of modes to solve and filter resolution. I would also appreciate any general tips that could improve the performance of 3D variability jobs.
I believe 512px exceeds the max box size for 3D-VA (at least in our experience). You will want to downsample prior to 3D-VA to a box size <400px, which will also make it a lot faster.
Given that your filter res is 8Å, you might want to downsample to ~2-3Å per pixel. You can always reconstruct from the original particle images using the low level results interface, so it doesn’t lose you anything.
in addition, I would dilate your mask by 20A (the mouvement will be looked at inside the mask so you want to dilate to increase your search radius). 8A filter resolution is the mouvement you are looking for, perhaps too big for your protein, perhaps 6A would be enough or maybe 4.5?
When you display the job, you can ask for the full resolution.
@vincent , on the resolution I think you mean the opposite, right? Rather try going lower resolution (i.e. filter resolution 10, 12 angstroms) as this usually contains the information of subdomains moving. The manual recommends using a resolution that is lower (= bigger number) than the one reached by the last refinement, or the calculations will get lost with all the noise. Giving up to 4 angstroms of data is only useful if you want to see very fine movements (side chains moving, PTM, ligands…) for which case you’ll need very nice starting maps, friendly particles and a lucky mask - at least this is the way I reason, but my stuff rarely go to better than 3 angstroms so I don’t even try high resolutions for 3DVA. And I tend to run unmasked 3DVA first, unless I already have an idea of the movement that is happening, for instance from 3DFlex or others. But yes, usually the outputs from the masked 3DVA job are nicer.
Anyways, just exposing here but I am not an absolute holder of truth, rather trying to develop the discussion.
@carlos I’m not the holder of truth either, it all depends on what is really available in the particles.
I usually test several resolutions but it is to get a little better here or a little better there, in my hands it is to get a better definition of the movement but it’s not going to revolutionize or see new details.
