Symmetry misalignment for icosahedral particle (cryosparc v2.15.0)

I am having a weird problem when trying to refine an icosahedral particle. The initial volume was reconstructed in Relion without any issues, and comes aligned to sym axes. After the very first iteration the symmetry becomes mismatched, so that 5-fold axis moves close (but not precisely) to where the 3-fold was. It seems that the system is rotated by exactly 90 degrees around z (the 2-fold axis) at some point.

This happens whether or not symmetry alignment is toggled.

I suspect this is a bug.


Argh! I see now: Relion and Cryosparc use different axes alignments for icosahedral symmetry. Both have Z along a 2-fold axis, but X and Y are different, with a 90 degree rotation between the two coordinate systems.
When the input model is aligned according to the cryosparc convention, the refinement proceeds normally. The result is not better than what we get in Relion, but I am happy it all makes sense now.