Rebalance orientations survey question

Cryo-EM Data Processing
1

Question for other users - if you found rebalance orientations helpful, was a reconstruction sufficient, or was a new refinement necessary to see an improvement?

IOW, were the orientations accurate, but too biased, or did the bias actually hinder alignment?

2nd question, were the improvements in resolution and map quality, or really just map quality?

Thanks :heart:

PS 3rd question: random vs. alpha vs. 3D error? My initial tests show improved resolution from alpha and a reduced bad/good conical FSC spread with 3D error.

3 Likes
2

  1. yes in some cases rebalancing orientations helps but it depends on tweaking Number of orientation bins, Rebalance percentile (usually 5 to 90%), and intra-bin exclusion criterion (alignments3D/alpha, random, pick_stats/ncc_score, alignments3D/error, alignments2D/error).

    1. alignment seemed ok, but I think the HR-HAIR could be good to try for accurate per-particle alignments in some cases.

    2. Doing the reconstruction right after is a good quick check. After the Rebalance 3D orientation job I would run the refinement (homogeneous or NU) with these particles for a apples-to-apples comparison.

    Overall, there are slight changes to cFAR/anisotropy, you certainly have to try adjusting several parameters with Rebalance 3D Orientations. Would agree that there are these differences between 3D alpha vs. error.

    https://www.biorxiv.org/content/10.1101/2024.08.08.607122v2 this was an interesting take CryoSPARC + RELION.

2 Likes
3

One other thing to consider: severe orientation bias has a more significant effect in ab initio and early iterations of a refinement (when not all particles are seen) than later iterations of a refinement. In some as-yet unreleased work following on the HA Trimer case study I found that starting with a good (i.e., isotropic) initial volume (still lowpass filtered to 30 Å) “rescued” particle stacks which otherwise produced hopelessly anisotropic maps (cFAR < 0.1).

So it may be that performing rebalance orientations early on to produce an isotropic map (albeit at a lower resolution) and then going back to use this good starting volume on the full stack may yield the best results.

As for your question about the modes – I haven’t seen one mode perform reliably better than the others. My usual caveat is that if your map is anisotropic, particles which match it very well (high alpha, low error) may in fact be worse than particles which don’t (low alpha, high error). I therefore tend to use random mode myself.

3 Likes