Indeed, I saw from the code that the PM trajectories are readjusted to the first frame for reference.
Also, I have since done some of the reading that I should have done before thinking out loud(!), and found out that in the absence of the -FmRef flag, MC2 uses the central frame as its reference–not the first frame. I’m not entirely sure what RC does but I assume the same given their shared provenance? Also, that BPP takes this into account when interpreting particle coordinates?
From the MotionCor2 manual:
The output and log files list the shifts relative to the first frame. However, the correction is
relative to the central frame by default. “-FmRef” is a switch that allows to choose the
reference either the central frame by giving it a non-zero value or the first frame by setting
it zero.
And for completeness, from this post, I got the impression that in PM all frames get shifted to some degree, i.e. there is no strict reference frame.
To clarify my thought process at the time, for others reading. Due to different frames of reference, the averaged micrographs from one will not line up perfectly with the other’s. So coordinates (x,y) from one may coincide with (x+a,y+b) from another. This amounts to errors in OriginX and OriginY going between the two. This is independent of the trajectories. Not sure if this rationalisation is how it actually works.
I take your point. gsFSC=0.143 is within 0.1Å. The caveat being that this particle set is limited by conformational heterogeneity so it’s been difficult to gauge. I would normally chalk it down to random variation and think nothing more of it, save that RC/MC2 has consistently (n=4 separate BPP+refinement runs) given slightly better FSC (across the resolution range, not just at 0.143) vs PM (n=3), so I thought to make a note of it. Internally, the RC/MC2 refinements have been rather consistent. As have the PM refinements. The ±line95 test was also within 0.1Å.
Another dataset where BPP has had more potential to improve the map may be a more suitable test case.
No, I haven’t in this case. The particle stack was imported directly into cryoSPARC for refinement. I thought less manual manipulation would be a fairer comparison.
EDIT: It’s probably worth mentioning that as a matter of routine, I perform preprocessing without a defect file.
I can run these tests. But to answer your question, I haven’t experimented with recentering. Also, a 2nd bootstrapping run would be using an improved reference. It may give a better result based on that alone. I should reiterate again that the choice of this particular dataset is a bit unfortunate. It is what we had available. I’ll check to see if one of our other EMPIAR cases gave a better return from BPP.
Cheers,
Yang