I am running heterogeneous refinement using particles extracted and Fourier cropped to a 64 pixel box. In my opinion, the image size should override the default 128 pixel box size. My job completed and I now have maps with 128^3 voxels and an FSC resolution estimate of 5.77Å. This is problematic considering my input images had a pixel size of ~3.3Å
I don’t think this is problematic - interpolating to a larger box size doesn’t seem to have any negative effects, and the resolution estimates out of heterogeneous refinement make sense to me when compared with density features … if you have an FSC of 1 at Nyquist and interpolate to a larger box size I would expect the FSC to extend a bit beyond physical Nyquist, no? Also I’m not sure anyone uses the resolution estimates out of heterogeneous refinement for anything in particular..
1 Like
You make some good points. Perhaps I’ve spent too much time reading 3DEM archives and am being a bit too rigid. I’m open to changing my mind about this.
I’m still not fully convinced this is the best default behavior though. In my example the difference in the Nyquist and estimated resolution is small but would oversampling the refinement box still be helpful if I had changed the default to something like 256 or cropped my particles much smaller?
At the very least is this not computationally inefficient?
I can tell you that it is much easier to interpret a 128px map than a 64px one from the same particles at say 6 Å in your example, and it is easy to change the setting if you want… I’m also not sure the run time is that different, but easy enough to test
I agree about the interpretation. I cloned my heterogeneous job and reran it with a box of 64 pix. At 128 the run time was 2hr 25min vs 1hr 35min at 64 pixels. Roughly the same number of particles in the main class but it might be interesting to look at the consensus particles between the two jobs.
1 Like