RBMC: pre-check particle distances (request)

Motion Correction
1

Hi CryoSPARC team,

For when you update Reference Based Motion Correction, may I add something else to the wishlist: a first-step particle position check-and-fail. Or, a particles-too-close-auto-remove option?

Also, can you clarify what RBMC considers “too close” and/or “duplicate” particles? Many people seem to be running into the same sort of issue.

Two RBMC runs fail at approximately 50% complete (6 hours of running) and 20% complete (2 hours of running) with a third at 75% and not failed (yet!) last night. I’ve got multiple distinct protein complexes from a single dataset, all resolving (so far) to 2.5-2.7A and all clearly different. Rather than carry out a single RBMC run for efficiency, I am having to run them individually or RBMC (eventually!) crashes due to “duplicate particles”.

If duplicate particle removal during 2D classification is functioning (and it does seem to be from deliberate tests) it beggars belief that after multiple rounds of 2D classification and then further duplicate particle removal set at 35A (done after the first RBMC crash, although now I’ll check larger distances) that there are still duplicates hiding. The FSC curve is healthy and falls cleanly to zero and stays there - when I deliberately test by adding duplicates, with even <0.5% duplicates hiding in the dataset it is absolutely clear the FSC does not reach zero.

Further, as RBMC removes micrographs with two or fewer particles, running RBMC on a per-structure basis is costing me quite a lot of particles from each set as many micrographs may contain 30+ “good” particles, but when broken down into distinct structures there might be only two…

Exactly what impact losing these particles has compared to optimising individual particle motion remains to be seen. When I know I’ll add an update.

Thanks in advance.
R

1 Like
2

As I said I’d update: the loss of the ~1,500 particles from the “two-or-fewer-particles-per-micrograph” cutoff did not appear to have an appreciable impact compared to the improvement from RBMC, as resolution went from 2.55Å to 2.3Å… :slight_smile: best case, but even worst case still saw an improvement.

3

Hi @rbs_sci,

Sorry this issue has turned out to be a significant hassle. We are planning to adjust reference motion so that it skips affected movies rather than crashing. This change will be coming relatively soon.

Harris

2 Likes
4

Cheers @hsnyder! :slight_smile:

Look forward to it. :smiley:

While a skip is a workaround, a split into subset option (in event of excessive particles) would be better from the user perspective (although significantly harder to implement I think). I’m experimenting with a dataset at the minute where particle count per micrograph ranges from 1,800 to 3,200… with <600 micrographs, dumping mics with higher particle counts might be costly!