Large difference in scores between v9.07 and v9.11

Hello-

I am trying out the latest version of Frealign (9.11) on a helical reconstruction of actin filaments in complex with a binding partner. I had previously run a refinement in v9.07.

Starting from my final refinement parameters with v9.07, I get an odd behavior where about 60-70% of my segments see a precipitous drop in their score from the ~20-35 range to ~0-2. This is accompanied by a corresponding decrease in the resolution of the map (both by FSC and inspection of features compared to known xtal structures) as these segments, which seem to have been contributing meaningfully to the reconstruction, are down-weighted.

There is no obvious connection to the scores of the particles from 9.07 (e.g. some very high scoring particles get a score near zero, while some medium scorers stay medium), which filament they are derived from, or something obviously pathological like very high shifts.

Interestingly, if I run the refinement setting the mask to "0 0 0 0 0" (e.g. just calculating scores without allowing refinement) I get the exact same pattern of scores, both starting from my 9.07 parameters and my initial parameters, which were calculated with an in-house SPARX IHRSR.

I have tried messing with PBC, which changes the scores slightly as refinement proceeds but not which particles have a high and low score. Also, I noticed that the mparameters indicate that a low stiffness now corresponds to a strong restraint, whereas the opposite used to be the case. I want no stiffness restraint, would "0.0" still be appropriate?

Also, keeping an eye on mrefine's log files in scratch, there is no obvious connection between a particle's CC and the resulting score.

I will continue trying to diagnose this myself, but any pointers in the right direction would be much appreciated!

Thanks,
-Greg

P.S. A v9.11 reconstruction from the v9.07 parameters looks great

P.P.S. The speed up and easy of use improvements are also great