I hope everyone is doing well. I am receiving the following warning on Non-Uniform Refinement jobs from particles sorted using 3DVA cluster mode. The warning only appears on some of my jobs (2 out of 6 total).
Set A is greater than set B by 2462 particles (2.25 percent difference relative to the total dataset). This is a difference of greater than 2%. If equally-sized Gold Standard splits are desired, please use the ‘Balance half-sets’ mode in Particle Sets Tools. Alternatively, ‘Force re-do GS split’ may be enabled, but this might not preserve Gold Standard independence.
I am wondering what is causing this warning and whether it is a serious problem that I will need to address before model building.
Your original refinement, which you fed to 3DVA, most likely had a very close split between particle half-sets. After 3DVA has run, and you’re refining the output clusters, the half-sets are not reassigned (this is for safety - again, as it says, to preserve the GS-FSC) but it’s warning about the imbalance. There are more particles from subset A than subset B in that 3DVA cluster, and the difference is great enough that it might potentially impact your results.
Occasionally, the imbalance might be great enough to completely ruin reconstructions where the original half-set splits are maintained. This would manifest as bad results from refinements and, more directly, one half map being “good” and the other “bad” due to the particle imbalance.
You can try rebalancing the split, but it may result in an FSC curve which does not reach zero completely. In limited (deliberate) testing, I’ve not seen this happen yet, but others may have done. All being well, the FSC will remain healthy and you might squeak out a fraction more resolution. Again, in limited testing, I’ve not seen an appreciable difference - but I’ve not tested extensively.
This warning was added because of a thread on the forum, which I’m currently failing to find…
