Hi,
In most cases, it makes sense to treat per-particle CTF refinements independently - in a perfect world, there wouldn’t be any correlations between defocus errors of adjacent particles, as they would be purely due to relative z-displacement in the ice layer.
However, I think there are a couple of common situations where a spatial prior would be worth considering:
- Small particles, especially when tilted. Small particles at moderate resolution often don’t have enough signal for per-particle CTF refinement to behave reliably, and when tilted I suspect that errors in defocus may be mostly due to errors in initial tilt/landscape estimation in patch CTF (e.g. due to the presence of gold, junk, whatever). Jointly refining per-particle CTF estimates with a spatial or tilt-based prior could help correct this.
- To compensate for global errors - e.g. where a gold edge or lump of junk causes an artifactual peak or valley during Patch CTF, throwing off CTF estimates for nearby particles; Another such situation is where the grid has a graphene or graphene oxide substrate, such that all/most particles are adhered to one interface. In this case, even for large particles, it might be helpful to perform an initial round of local CTF refinement with some kind of spatial prior, followed by a second round without it.
Cheers
Oli
EDIT: This is supported, I think, by observations here - where Patch CTF improved resolution for tilted data vs Gctf, but per-particle CTF refinement further improved resolution at higher tilts (by 0.4 Å in some cases). It seems likely then that for small particles at higher tilts, adding a spatial prior could help matters further.