I was looking around to see if this has been asked here before and Gemini/ChatGPT didn’t give me good answers, so here we go.
I have a dataset collected on hexafoils (pix size around 0.7 Ang pix size) with a relatively small protein at 100-90 kDa (about 100 Ang in the longest dimension). The protein is sort of oblong diamond shape, the overall shape is a bit like our favourite enzyme Beta galactosidase but of course smaller. I’m in pharma so I cannot share the identity, apologies.
Currently, I mostly see only a few orientations which I simply think is due to the size of the particle. In the past with similar small proteins, Topaz has been really helpful to tease out rarer views.
However my main issue is that I cannot seem to find parameters that allow me to use Topaz without just picking on the gold. And I really mean, it will mostly JUST pick on the gold.
I’ve tried training, cross validation, and even just extract with some random previously generated models (either from other refinements or the standard built-in ones).
Anyone that has a good workaround to let Topaz pick at least 50% in the wholes or ideally better than 90%?
I have not used topaz on data from gold foil, so can’t advise on this.
But have you tried the “standard” picking procedure in CryoSPARC? It has become very good thanks to the micrograph denoiser and junk detector. And now in v5, if you run the junk detector just after motion correction, the resulting junk masks will help the Patch CTF and Micrograph Denoiser jobs (examples are shown in this talk).
Try the blob picker on denoised and junk-annotated micrographs, then run the Inspect picks job to filter the resulting picks, and see what you get. I have not tested rigorously against topaz, but my impression is that this method is on par with it for many datasets.
I immediately tried to test it out but I’m not getting anything similar to the micrographs that Ali showed in the presentation - junk detected and denoised. I wonder if you would have a simple “how to” with the basic steps and procedures?
My understanding from the talk and previous use of CryoSPARC is that it should “just work” if you run the junk detector after motion correction and run the rest as usual downstream of the junk detector. I just tried it on a few micrographs and everything ran normally when connecting the “labeled micrographs” output slot of the junk detector to the input of Patch CTF, and the output of Patch CTF to the input of the denoiser. There doesn’t even seem to be a switch to toggle in downstream jobs to tell them to use junk masks.
But the visualizations shown in the talk don’t seem to be generated by the downstream jobs, so I agree with you that it is difficult to tell whether the junk annotations are taken into account. Maybe someone from the team can confirm?
I may be wrong on this, but my impression is that the current workflow is that pickers cannot use junk information. Based on the how-to page ( Job: Micrograph Junk Detector (BETA) | CryoSPARC Guide ), I think that you have to run junk detector and if you want to filter particle picks, you can re-run junk detector after picking particles and it will remove any close to the junk masks, but it doesn’t use junk information during the particle picking process.
If you’re using hexaufoils and have lots of gold in your images, you may also want to check this thread out – your processing may be limited based on the amount of gold in the image, depending on the size of your camera (4k x 4k at 0.7 A/px likely will have too much gold, other camera configurations may not).
Hi all, thanks for posting and the interest in the new features that are upcoming in CryoSPARC!
Just to clarify in terms of what has been released:
Micrograph Denoiser was released in v4.5
Micrograph Junk Detector was released in v4.7 (and can be used to identify junk, and to discard particles close to junk)
Junk Masking (i.e., where junk masks are used to mask away gold and other junk during preprocessing and picking) is not yet released, but will be in the next CryoSPARC version (v5.1)