I am working on a membrane protein and using the new NU refinement I get a weird tendency that I can’t get to overcome by changing parameters. The dataset is relatively small, around 90k particles, but the protein is a homooligomer with rotational symmetry. During the initial rounds of new NU-refinement the computed FSCs go down with each iteration and with the auto-increase of the batch size. However the FSC filtered Side A/B graphs appear noisy with noticeable rotational streaks. After iteration 2-3, and especially when using the full dataset, the computed FSC goes significantly up (from 6 to 8-9 A) instead of continuing to go down and the final corrected FSC curve features the notorious sharp dip with resolution worse than that on the ‘no mask’ or ‘circular mask’ curves. The same dataset refines to about 7A using the legacy NU refinement with C1 and although there is a dip in the corrected FSC curve, it is pretty much negligible and far above the 0.143 cut-off. I did refinements with and without symmetry, changing the dynamic mask threshold, increasing the latter’s near and far parameters, enforcing non-negativity, etc.
Has anyone observed such a weird behavior with the FSC estimates during New NU-refinement and especially noisy/streaky reconstruction plots in the intermediate steps. Any suggestions of how to overcome this?
Thanks in advance,