3D classification fidelity in cryosparc


I wonder how the fidelity of Cryosparc 3D classification (heterogenous refinement?) compares to Relion 3D classification. By fidelity, I mean the tendency of particles to cluster together after multiple identical runs of the classification.

I have seen and read many lectures / discussions regarding this issue from Carlos Sorzano and others, but I continue to see the number of particles / inferred free energy from the number of particles being reported in the literature from either Relion or cryosparc without regard to this issue.

The assumption of course is that, in those promiscuous classes, there must be a homogenous subset of particles (could be only 30% or 40% of total images in the class) that give the high resolution details. The rest of the images will behave as noise in this case. My understanding is that by enhancing the fidelity of classification, we can achieve a reproducible and consistently higher SNR volume of the same promiscuous reconstruction. Also in that case, the relative number of images, provided that is consistent and correct, could be a direct indication of the relative free energies of the different class structures.

Has anyone experimented with Cryosparc enough to give some thoughts or suggestions on the matter?!

1 Like

Hi @ahmadkhalifa,

Circling back to this post after a few months! Have you tried out our new 3D Classification job? If you (or others here) have any updated thoughts / feedback on your experience with multiple runs of 3D classification, class robustness, and any comparisons to heterogeneous refinement / RELION classification, we’d love to hear them.