Dear colleagues,
I am searching for advice on verifying whether my protein has D1 or D2 symmetry. X-ray data suggests D2 (a squarish shape - dimer of dimers, maybe), but the images and ab initio reconstructions leave me uncertain.
Could you share experience or recommend reliable image processing methods to confirm D2 versus D1 symmetry?
Kind regards,
Dmitry
In cases like these, I usually do everything in C1 to see what resolutions I can get to, then start testing different symmetries - it’s usually pretty easy to see what sort of symmetry should be applied, and if increasing too high it’s fairly normal for the map to start looking really messed up. So try both, and see if one or other map looks broken.
Beware, too, pseudosymmetry! We collected a dataset last week which looks like it should be D5, but has an asymmetric core which gets absolutely wrecked when any symmetry is imposed.
Thank you very much, @rbs_sci, for your excellent suggestions and prompt response! I should have mentioned earlier that I’ve already tried C1, D1, D2, as well as refinements and following comparisons, including D2 pseudosymmetry. The pseudo-D2 seems more natural, although, as expected, it’s a bit lower in resolution than pure D2.
I was wondering if there might be an algorithmic decision tree—including a symmetry-defining tool (as the one existing for helical analysis) — that could help analyze (pseudo)symmetrical maps (in future developments of CryoSparc maybe?). Running through such a sequence could provide clear proof that the algorithm can reliably distinguish between all possible options, including pseudo-symmetrical ones.
Thanks again—I’m glad my approach aligns with your recommendations! 
Kind regards,
Dmitry
The problem with trying to automate symmetric processing it is that automation is pretty bad at (quickly) figuring out the “quality” of the map - something which is immediately evident with a simple eyes-on check to a user. At least unless it spends a lot of time on it.
A “simple” check would be final resolution (likely also including cFSCs), which may or may not get better or worse as higher symmetries are imposed (see also my recent post about poor alignment of initial models messing up results). If imposing a wrong symmetry and the resolution gets worse, then the program may interpret that correctly (symmetry wrong) but if resolution improves but map quality worsens, it may interpret that incorrectly as the “correct” symmetry.
A more in-depth automated assessment might include, at least how I imagine it, Model-Angelo de novo model building into each map (with both hands), with and without sequence, which could them be assessed for, e.g. number of chains built, length of chains, how similar the chains are (effectively Chimera’s MatchMaker followed by an RMSD check/report). The idea being that a C4 or D2 symmetric protein would have four chains with low RMSD, C2/D1 would have pairs of chains with low RMSD but incorrect pairs would be high RMSD. The issue there is that would take a fair while even for smaller maps… for larger maps/volumes or those with many chains that will take a very long time! And to be truly automated it would need to try all symmetries you would need some way of telling it “Stop! Don’t try any higher symmetries!” or “only try these symmetries” (which removes a large part of the automation). Further, chain checks on complex structures (e.g. photosystems) might get the algorithm confused if too many mismatches, so how the chain checks were implemented would have to be done very carefully…
To do fully automated symmetry searching reliably, the map would also have to be rather high quality even with no symmetry imposition… I’m sure it’ll be solved in the future, but it would still be an assistant, rather than a replacement for the researcher. 