After running NU refinement (cryosparc v4.6.2 - I have not gotten around to updating), I get odd looking halfmaps while the final refined map looks OK.
Looking at the histograms in chimeraX there seems to be a filtering/scaling problem. Previous posts have discussed issues where the Nyquist of the extraction box far exceeds the refinement resolution I am not sure if this is the case here as in the previous posts refinement resolution were at least 3 x Nyquist. Here I have (pixel size is 0.9202 Å, Nyquist 1.84 Å).
If I filter the halfmap in chimerax using volume gaussian #6.1 sdev 1 - I get the following, which looks more like what I would expect the output to be:
This appears to be some sort of bug in outputting the halfmaps. My first instinct was that these halfmaps looked sharpened? How are halfmaps treated in cryosparc? I don’t think these can be deposited to the pdb/EMDB as are. Is there any work around? or way to re-output the halfmaps?
I should note the outputted halfmaps with the odd looking histograms does not appear after every refinement (similar FSC resolution and Nyquist values). Earlier refinements looked either OK or like above. But I cannot see any trend which would trigger it. I had originally thought it was related to RBMC or CTF aberration correction but some jobs outputted halfmaps with histograms consistent with the final volume, others not.
Your halfmap looks like what i would expect for a local refinement for the part outside the mask. Maybe check the masks (mask_fsc/mask_fsc_auto/mask_refine).
Your levels for the halfmap are off, that from the standard deviation for before? The histogram shape looks as expected.
Grab the particles run through homogenous reconstruction job and see if that gives the same thing for half maps.
NU refinement was ran with dynamic masking start resolution set to 1Å
The mask_fsc_auto and mask_fsc look as I expect them. The refinement mask has levels 1 -1 - and it looks like it is the extent of the box. Which is what I would expect as I have disabled dynamic masking. This refinement mask strategy is also used in jobs that output halfmaps with histograms similar to the refined maps.
“Your levels for the halfmap are off, that from the standard deviation for before? The histogram shape looks as expected.”
I am not sure I understand this. Do you mean that the levels are off for the halfmaps before or after I run the chimerax command? For me the histogram and levels for the halfmaps from cryosparc are odd. I would expect the shape to be similar to the final refined map, as is the case for other refinements ran similarily
Running Homogenous reconstruction gives similar looking noisy half-maps with histogram profiles as indicated in my original post. FSC mask looks as expected.
This happens occasionally with NU refine. Never worked out why. Don’t think the Structura team have either. Hadn’t seen it for ages, but a new student triggered it last week on 5.0.1.
Interesting and many thanks for sharing - it is also random in my hands - I see no correlation with particle number, particle handling (CTF correction, RBMC) even in the same dataset. Sometimes its fine - other times it is not.
It can only think that some sort of normalisation is not happening - especially when I compare to the final map. Beyond I guess the why (which I am still interested in) - my overriding question is can these halfmaps be deposited?
That I haven’t tried. I guess I need to look at csparc2star.py this evening! Last time I tried this was from cryosparc v2 I believe. I guess the same workflow applies. The data is separated into exposure groups.
Should be straightforward, only thing to remember is that higher order CTF params (e.g. beam tilt) will not transfer, so if those are important in your case you will need to re-refine them with CtfRefine in Relion
Whenever you are inspecting half maps from CryoSPARC, it’s good to remember that we perform no filtering whatsoever – they are exactly as reconstructed from the particle images. As I’ve mentioned in the posts you reference@Joe, unless your reconstruction is very close to Nyquist, the noise from high frequencies will swamp out the signal in your unfiltered half maps. Keep in mind too that, when looking at the whole box like this, the unmasked FSC may be more useful in gauging your expectations.
You can see this effect yourself by taking a reconstruction which has not reached Nyquist (like this one) and comparing its half maps to half maps produced by reconstructing downsampled images (downsampling is, in essence, a lowpass filter). The particles used for this diagram are the exact same particles with the exact same poses, but the lower row is a reconstruction for which the images have been downsampled such that the reconstruction is much closer to Nyquist.
I’m happy to answer any more questions you all have on this topic!
Edit because I forgot to close the loop: in general, the appearance of noise in half maps is expected and is not alone reason to worry about your reconstruction.
