Question nice 3DVAR intermediates, but lower resolution refine

Hi cryosPARC devs/users!
I obtained some very nice results disentagling variability with 3DVar after playing a lot with it, and Im really happy with its power: I reproduce some conformations that I previously capture on Relion, but 3Dvar allowed me to resolve some new very interesting (that also got biological sense). When performing 3Dvar with intermediates and outputing the each frame of a series, I noticed that the reconstructions are quite good. Their details agree with the number of particles in each frame. The issue that I cannot solve yet is that when refining the outputed particles of each frame independently, I obtained maps that look evidently lower resolution that the corresponding reconstruction from the 3dvar-intermediate job (many details that got sense in the volumes series are lost: overall each refined set of particles is “identical” to the volume output of that frame, but way lower res). If I understood correctly, each volume.mrc for each frame in a series is a CS reconstruction from the particles on that frame (so I was expecting that when refining each set of particles I should obtained at least a similar map). May be I am missing something really evident?
(Briefly, my path was performing 3Dvar with intermediates and output particles, then refine each set of particles of each frame using my consensus refine as ref (rolling window 0, tophat). If I compare each refine map of a series, with the corresponding reconstructed map of the series, they are all systematically worst…despite shape and movement is the same. I try several things, homogeneous, heterogeneous, extending number of passages, LPF, but I couldnt reproduce the “quality” of the map cs gave when reconstructing intermediates). Again, may be Im loosing a key point here, but I appreciate any insights to figure this out. (I can give further details if needed, but keep it like this now to not extend more)
Best and thanks a lot in advance.

Hi again,
just as an update, after refining quite a lot of the outputted particles per frame from the 3dvar/intermediates job (and trying different parameters) I consistently see that the cs-reconstructed volume in all series appear significantly better resolution that the corresponding output particles. Surely the effect is most notorius in in frames with less particles where the cs-reconstruction is way better that the refined map. What can be the cause of this difference? I would have expected a similar result at least, or better, from the refinement than the backprojection. I would like to improve the refines resolution at least to the level of the reconstructions as the movements captured by 3dvar were indeed way better than other strategies I perfomed . @apunjani may be theres something conceptually big I am missing?
Thanks a lot in advance for any comment!

Can you reproduce the subset volumes from 3DVA display with a Homogeneous Reconstruction Only job, enabling Override FSC Filtering and specifying Filtering values as used for the 3DVA Display job?
Out of curiosity, what job type(s) produced input particles and mask for your 3DVA job?

Hi @wtempel,
thanks a lot for the reply! Yes, I can reproduce the volumes under the job/conditions you suggested, and again the resulting map looks sensibly lower resolution. More precisely, the map looks identical as I obtained before using NUR (i did several trials, and NUR gave a bit better results, so I stick to it).
I remark that this effect is more notorious in frames with lower particle count: here the reconstruction looks way better that my refine. When the particle count is higher, they tend to be similar despite the reconstruction still looks better.
My input particles and mask come both from relion (and as a sanity check, before starting playing with 3dvar, the particles refined -using the mask- correctly in CS). Is the reconstruction of intermediates using a Homogeneous Reconstruction Only job (say with the default parameters) on each same groups of particles that are outputted? (that was my understanding if using a tophat rolling window, if not mistaken)

Hi @wtempel
Just a new update. After many trials, the combination that gave a map that looks the closest in similarity to the one from the 3dvar-intermediates reconstruction is a HRonly in which I replace the alignments3D in the low-level for those of my consensus refinement (the one used for the 3dvar analysis)…(That is, I refine on the intermediate output particles, but replaced the particles.alignments3D only)…and override filtering resolution 12 (I got 20 on 3dvar, values other than 12 rapidily make it worst). Not using my consensus refine alignments gave me always sensibly low res looking+not detailed maps.
So Im more curious now about what parameters use the reconstruction performed during the 3dvar-intermediate job, specially cause with low particle content the reconstuctions still look way better
Is it only using the particles_series_frame data? My thought was that it’ll be a backprojection HRonly job on each group of particles (also, I used a tophat windowing).
Thanks a lot in advance!

“Intermediate” reconstructions during a 3D Variability Display job are equivalent to Homogeneous Reconstruction Only using the given “frame”'s particles and their alignments, which are unchanged from the alignments submitted to the 3D Var job.

1 Like

Hi @wtempel
thanks for the reply. ok, so my understanding was correct then. I played afterwards with some different parameters and was able to reproduce the results. Which was also a very illustrating exercise btw.
Besides, still getting impressed by the power of 3dvar to disentangle heterogenity (in my case, which is tricky, its helping me a lot to shed light on it). So happy with it.

1 Like