Hello folks, I am trying to understand some problems arising from Homogeneous Reconstruction jobs, and subsequent refinement jobs on these particles. I am testing HR-HAIR (thanks @olibclarke !) on my 70kD particles, which works incredibly well for 2D classification and Ab Initio, but later steps are falling apart for me and I want to diagnose the issue.
58k particles were taken from a good Ab Initio class, then either left alone or re-extracted with a smaller box or without binning, and reconstructed using the Homogeneous Reconstruction Only job. While the output volume looks highly similar to the Ab Initio volume, the GSFSC raises concerns.
-
Binned particles (384→192) produce GSFSC curves that never really drop, from Homogeneous Recon.
-
Unbinned particles (384) produce seemingly overfit volumes in Homogeneous Recon, but GSFSC looks okay.
-
As per HR-HAIR, Local Refinements seem to work on the volumes produced from Homogeneous Reconstruction (despite the GSFSC concerns), but exhibit poorer resolution compared to the Homogeneous Recon.
-
All other refinement jobs drastically decrease the resolution of the Homogeneous Reconstruction or the Ab Initio volume, often decreasing it all the way to 7A and losing all detail
Is there something wrong with how I am implementing Homogeneous Reconstruction or subsequent refinements? If anyone has insight or resources, I’d greatly appreciate the help!
PS - There is a very similar discussion (Homogeneous and NU Refinements are worse than Homogeneous Reconstruction?) on this, but I believe their result was duplicate particles, and I am confused why my 2x binned particles don’t make GSFSC go to 0, but unbinned particles do.