Hi, I’d like to do local and global CTF refinement of my particles, which come from a dataset collected using a 3x3 beam image shift (i.e., 9 exposure groups for global CTF refinement). Can I simply take the 9 exposure groups and input them into a non-uniform refinement job with both “optimize per-particle defocus” and “optimize per-particle defocus” turned on, representing local and global refinement, respectively?
The following CryoSPARC documentation makes it sound as if this is not possible, but it seems to be doing exactly that.
Non-uniform refinement with high-order CTF correction
In CryoSPARC v2.12+, the Non-uniform Refinement job has been updated to use the new GPU code that supports higher-order CTF correction, but this is NOT enabled by default. You must turn on the Enable higher-order CTF parameter in Non-uniform refinement. Please also note that legacy refinement jobs will not support the correction of high-order CTF aberrations or anisotropic magnification.
On-the-fly CTF refinement cannot be done during a Non-uniform Refinement, so particles should be processed through the standalone Local CTF Refinement then Global CTF Refinement jobs first.
Thanks,
Andy
The information you quote is quite out of date, as CryoSPARC 4.7 is rather different to CryoSPARC 2.12… 
Use Exposure Groups to split your exposure groups (if not already done so). Then feed that to NU refinement.
I generally do not recommend doing higher order corrections during NU refinement, as it does not limit iterations for it, and so, sometimes the beam tilt parameters will explode (particularly if enabling multiple higher order corrections at once) and the refinement will “succeed” but the reconstruction will be a ball of nonsense with higher order parameters “converging” at millions of milliradians - a darkly amusing idea that the beam is doing loop-the-loops down the column. Defocus refinement is safe, though. However, because of this occasional parameter explosion, I now do higher order parameter refinement only via Global CTF Refinement; multiple runs, usually, tuning each parameter one at a time, often with NU refinement interspersed for a sanity check.
1 Like
@rbs_sci have you done any systematic comparison of the order of the CTF refinements or if alignments should be updated in between? If it is all avoiding these parameter explosions it makes sense, but otherwise I don’t understand why separating would help since the component refinements are done sequentially anyway.
(I.e. as I understand it there should be no difference between doing 1 iteration of Global CTF refine with all aberrations on, and doing 1st one with Zernikes and 2nd one with aniso mag).
@DanielAsarnow I did a few tests when I saw it happen originally across different samples but nothing exhaustive. Aniso mag (used to?) require two passes, and if turning on tetrafoil refinements they would sometimes become excessive on the second pass, which is why I split aniso mag and tetrafoil to independent runs.
For a long time I did one-parameter-CTF-refine/refine3D/next-parameter-CTF-refine/refine3D/etc in RELION because I was seeing notable resolution improvements through the improved angular assignments versus dumb reconstruction - but in the end it turns out to be highly sample dependent: the larger the particle, the greater the impact…
I agree on your last point; if things are pretty well estimated already, maybe no worries. If one is a bit off, I think it’s safer to refine one at a time in case it corrects the better optimised parameter against the one with greater error first, then snowballs. But that may just be me being over cautious.
I found defocus refinement → aniso mag → beam tilt/trefoil → aniso mag → defocus refinement → anything else was best for general application.
But I should have a deeper look at this, if only for interest.
2 Likes