I am working on reconstructing a particularly tricky particle, and I thought this setting may useful, but i wanted to clarify what it does and whether it’s appropriate to my use case.
I am trying to reconstruct a globular protein that extends off of a larger parent particle. in reconstructions of the parent particle, the protein of interest is disordered, i believe for two reasons, the first being a small degree of motion, and the second being a symmetry mismatch between the protein of interest and it’s pocket that results in disordered density.
The description of “Use pose/shift gaussian prior during alignment” leads me to believe that particle images that undergo/require large rotations/shifts to align are down-weighted in the reconstruction; is this a correct interpretation? I was wondering if i could use this setting to down-weight particles of interest that have experienced a higher degree of motion and would thus require larger rotations/shifts to be aligned to the initial volume?
Apologies if I’ve drastically missed the mark!
Jobtype Local Refinement. I always assumed it meant that there is a lower probability that a particle would be reassigned to a large distance/rotation away in any direction. So they penalize any decisions to assign a large shift. But in the end all particles contribute the same to the reconstruction.
From the guide:
As an antidote to overfitting, we can regularize the alignment problem by constructing a prior distribution over pose and shift, in order to penalize alignments that are too far away from the original alignments. The implementation uses an isotropic gaussian prior, centered at the current best alignment parameters. This can be done by activating the
Use pose/shift gaussian prior during alignment parameter, and by specifying the
Standard deviation of prior over rotation/shift parameters.
These standard deviations should reflect the size of movement you expect to see. If you have a complex that undergoes a large amount of motion (e.g. a spliceosome head region, such as in our case study), you may want a larger prior like 20º and 10Å. On the other hand, if you are using local refinement just to try to improve detail in a specific region of the map and don’t think there is much independent motion of that region, much smaller priors, even as small as 3º and 2Å may be optimal.
I recommend you always activate these three settings:
- Use shift/pose priors
- Recenter poses each iteration
- Recenter shifts each iteration
IMO they should be on by default. Most users are surprised to learn that without “recentering” the same initial input angles are used again and again in each iteration. I also usually find that local refinement is not effective without recentering, but that it tends to diverge without the priors. OTOH if they are made the default I’ll lose the ability to look like a miracle worker when I advise other users…
BTW, to control the search, just change the angle and shift standard deviations but leave the search extents as the default (3 * standard deviations). I use e.g. 7˚ and 4 Å for refinements already around 2-2.5 Å and slightly bigger ones for lower resolutions. Probably the particles are not 10+ pixels off in shift if the resolution is already reasonably good.
The initial resolution is also important to consider. If you can already see secondary structures in the focused region, I would use something like 8-12Å. If the resolution is already ~3Å I tend to use ~6Å.
Thanks for the tips i will try enabling these settings and see if my results improve I have never used any of these settings but am very curious to see the results, running a few tests as recommended by yourself! Some of my reconstructions are at ~3.7 currently so using slightly larger numbers as you suggested.
I am worried that the main local reconstruction i was aiming to use this parameter for is perhaps a lost cause; the particles produce distinct but disordered density, we think not only due to motion but due to binding in multiple orientations (causing blurring when symmetry is necessarily imposed to the broader reconstruction). I have tried a lot of things including 3D classing to try and tease out the individual binding poses, but the protein seems to move a bit and is quite small relative to the central density and just doesn’t seem to be aligning well. But i will try these parameters and see if they help at all
Thanks for your time,
Thanks so much for your reply I understand my misinterpretation now. I will tinker with this setting i think and see if i can get some improved results. We do expect the protein has a reasonable degree of independent motion, however less so than other similar proteins (our particle is a viral decoration and other proteins are known to occupy similar environments). We suspect the disorder is likely more related to symmetry mismatch, however i also suspect the motion is not helping with alignment so this hopefully will be helpful
Likely you already tried this, but have had some success with things like cryoDRGN and/or 3dVA in cases that I think are like what you’re describing (large degree of motion of something small).
I have not tried these options but i really appreciate the suggestion; we have kind of tabled this issue for now as it was not really working, but we intend to have another go in the future so i will give these a go