Reconstruction only

I have nice little refinement at 2.7A with 1M particles of a 100kDa membrane protein in DDM micelle (in not so thin ice). I decided to redo 2D classes with the results of the local refinement, take the very best classes and do a reconstruction only, i still obtain 2.7A with 1/3 of the particles number so far so good. If restart a local refinement (same mask) with the subset of particle my resolution drop back to 2.9A.
So naive question, as usual, why is the refinement resolution worse than the reconstruction only?

anyone thoughts on this?

Hi @ehanssen,

There are a few reasons why resolutions may differ, including masks, phase randomization effects, marginalization, etc. For fully consistent comparisons, we always recommend using separate “Validation (FSC)” job on the half-maps of each of the jobs you would like to compare, while keeping the input masks and parameters constant. Could you launch two Validation (FSC) jobs with the same mask, one on the final local refinement refinement output, and one on the original refinement output, and see if the FSC curves differ significantly?