I was wondering what would be the best strategy in processing tilted and untilted data collected by AFIS…
Should tilted and untilted data be mixed since the beginning rather than processing them independantly before merging high resolution good particles ?
What is the impact of tilted data with 2D classification (with wrong CTF estimation per particle) ?
When is the right moment to correct for tilted data : just after first refinement doing Global ctf refinement with beam-tilt info followed by a local CTF refinement ? Or rather at the end when you identified a high resolution stack of particles ?
I usually run the two (tilted and untilted) dataset separately in bin 4 through two rounds of 2D classification and 1 or 2 rounds of heterogeneous refinement (with 2 good classes and 1 trap from a previous project). After this, I combine the remaining particles, extract in bin 2, and do further cleaning with heterogeneous refinement. I wait until the end to do global and local CTF refinements.
This is simplified for me a bit, as I have good references. I have difficulty generating ab initio models with data with limited views. As such, you might need to combine your particles sooner, to get a good ab initio model for down stream processing, unless some are already available in EMDB.
Conversely, I usually merge them ASAP. When I’m working with tilted data, it’s usually because untilted is completely anisotropic and next to impossible to do anything with beyond some nice (but nearly identical) 2D classes. If the data were collected independently (e.g. different EPU sessions) then I’ll import, motion correct and CTF estimate separately, but from there work with all inputs.
Patch CTF estimates locally across the micrograph, not a global estimate like CTFFIND4, so tilting doesn’t cause CTF estimates to be so far off the way CTFFIND can.
As for global CTF refinement, that’s something to do when you’ve already got a decent reconstruction. Global CTF refinement does better the higher the resolution reference you give it (unless you manually force fitting to higher resolution) so doing it early may lead to erroneous tilt refinements which can cascade from there.