Minimum recommended particles per micrograph for RBMC?

Hi all,

There’ve been a number of threads about the maximum number of particles possible based on memory constraints and how RBMC seems to be especially sensitive to particles that are close by to each other. However, is there a lower limit for the number of particles per micrograph that’s required for RBMC?

Since the spatial prior requires the trajectories of nearby particles to be similar to each other, it seems like the accuracy of this prior would decrease if there aren’t enough nearby particles to compare against. In the tutorial example, RBMC used ~11,000 particles from 20 micrographs in the “Optimize Hyperparameter” step, which comes out to ~550 particles/micrograph. However, in my own dataset, I “only” have about 100 particles/micrograph. Based on other posts on this forum, I think I’ll be ok. However, is there a lower limit below which RBMC can no longer calculate accurate hyperparameters for the spatial prior? Would this hold true for Bayesian polishing in Relion?


I had a sample with ~10 “good” particles per micrograph recently and the RBMC hyperparameter estimates and motion traces looked sane.