Dear CryoSPARC team, I have a small question regarding the average power spectra job. Apart from using the in-plane angles to align the segments vertically, calculating their PS after padding, and summing those PS, are there any other steps done ? For example, are there any attempts to better align the PS ? I have a strong signal in a PS sum that shouldn’t be there (I know the symmetry), and I’m trying to figure out where it could come from…
Thanks much in advance,
Ambroise
Dear @adesfosses,
Thanks for your question. No – as of the latest CryoSPARC version (4.5), nothing else is done – the particles in each 2D class are padded, PS computed, rotationally aligned, and summed together.
Could a possible explanation be related to the 2D class having residual out-of-plane tilt? We have received a feature request to allow 3D alignments in this job for that reason (although this requires a previously refined structure)
Best,
Michael
Dear Michael,
thanks for your clear reply. It must come from something else, likely a “contamination” of those classes by segments corresponding to another symmetry (in this case out-of-plane can’t cause this signal). Just wanted to check that there was no attempts to align the PS other than using the existing in-plane rotation.
With best wishes
Ambroise
Dear Michael,
are there also not any attempts of “normalizing” the output PS or compensating for amplitude decrease ? I often see stronger signal (white area) at high res than at middle resolution, which is not really as expected, as well as other unusual distribution of grey values across the different resolution ranges. Thanks for confirming 
With best wishes
Ambroise
Dear Ambroise,
Thanks for bearing with me again. The power spectra shown in the jobs Event Log are shown in logarithmic scale (rather than linear scale). The .mrc images of the power spectra written out to the job directory are raw values (not logs) of the summed power spectra across all particles in the given 2D class. Is it possible that this difference explains the relatively high strength of signal at high resolution that you are observing, if you are seeing this in the Event Log images?
The only other factors that I have yet to mention that are worthy of noting are:
- particles are windowed in real space according to the windowing parameters in the job;
- bilinear interpolation is used to rotate each image’s power spectra before summing them together.
It may be worth changing the interpolation order from 1 to 3 and seeing if the described artefacts disappear. Thanks for reporting and I hope this information may be helpful to discern what may be going on.
Best,
Michael
Dear Michael, thanks for your reply. I was initially surprised -in my second question- by how strongly the amplitudes rise at high resolution (see the image), but it is actually present in the raw images too, and seems to be characteristics of recent datasets (I guess this is the NPS ?). The only issue with this is that since helixplorer is based on the grey values of the pixels to calculate the score associated to a particular symmetry, it is harder to get a usable score with such PS. Maybe a solution, in addition to outputting the normal sum of PS, would be to calculate the baseline of the amplitude variation, and subtract this baseline in a “baseline corrected” PS ? Like in eq 1 of Influence of electron dose rate on electron counting images recorded with the K2 camera - PubMed ?
In any case, cryosparc did not introduce anything that was not in the data, sorry for bothering you !
Thanks for the help
Ambroise
Hi @adesfosses,
Thanks for bearing with me. One other thing is coming to mind that may explain the increased amplitude at high resolution. In motion correction, dose weighting is done to upweight earlier frames at high resolution, and to downweight later frames at high resolution. However, at low resolution, frames get a closer to uniform weight. This is described a bit more on our guide pages.
This means that at high resolution compared to low resolution, there will be a larger variance in the noise in the images. This may be the underlying reason why there is a larger amplitude variation at high resolution (the power spectrum you’ve shown growing “whiter” near the periphery of the box)
I was thinking something worth trying could be to average together particle images that have been extracted without dose-weights applied. If you completed motion correction in CryoSPARC, there should be an output result of motion correction called micrograph_blob_non_dw. It might be worth trying to re-extract particles with this result overridden using the low-level results interface (described by Rich here), and then re-run 2D Class and Average Power Spectra.
I’m not entirely sure this would solve the increased amplitude at high resolution issue, but perhaps it is worth trying. If it doesn’t resolve it, my apologies I couldn’t be of more help
Best,
Michael