Dear all,
When I connected the polished particles produced by reference-based motion correction job to NU-refinement job, NU-refinement got stuck in “running local cross alidation for A…”, and no progress was output even after 10 hours running. However, homogeneous refinement was ok. Does it take a long time to do regularization for NU-refinement after RBMC? or is there any unfixed bug?
The box size here is 880px. What was your box size prior to RBMC? 880px will indeed take a long time.
I suspect that you have run RBMC without Fourier cropping, perhaps on super res or high-mag movies?
Thanks for your kindly reply. The box size of input map is 440px with pixel size 0.98 angstrom, and the pixel size of motion-corrected micrograph is also 0.98 angstrom. I left Fourier crop to box size blank. According to the cryoSPARC guide of RBMC, “the particles are extracted using the raw pixel size of the movies (including the upsampling factor, in the case of EER movies)”, because the micrographs were EER format, the particles were extracted using pixel size of 0.98/2 angstrom. I will try RBMC with fourier cropping to 440px. Thanks again.
Rather than time-expensive RBMC’ing again, just downsample the output. 
edit: How close are you to physical Nyquist? If you’re getting close, downsampling to, e.g. 1.5x of 8K (so, 660 pixel boxes rather than 880) may yield a better looking map which is easier to interpret, even if resolution is unaffected.
1 Like
I tried RBMC with fourier cropping to 440px, the same pixel size with that of input particles and map. After that, NU-refinement worked and the resolution improved about 0.15 angstrom. Thanks very much.
The final resolution is about 2.7 angstrom, about 0.8 angstrom apart from physical Nyquist. You mean downsamle the output particles from RBMC and then run NU-refinement? I guess the inconsistence of pixel size between RBMC output particles and map (due to RMBC considering upsampling factor for EER format) is the problem in my case.
Correct, as downsampling is faster than re-running RBMC. And too close to Nyquist, maps can be tricky to interpret.
But if you already reran with fourier cropping, then downsampling is irrelevant.
Thanks. I got it. I noticed there is a downsample particles job in the cryoSPARC gui.