pyrosetta.rosetta.core.pose.rna

He! I am super surprised as I have no idea what that (ugly) setting of PyMOL that is… If you show sticks, you will likely see that the ribose ring on that odd one is open.

In the params files of rings, you see that there is a cryptic bond cutting entry —this is allows the bonding in rings to be described with a linear sequence of atoms without causing any algorithm to go round and round the ring looking for bonds. However, if the strain is too much this cut bond will fail to close.

Minimising it and its neighbours with FastRelax under cartersian settings  (minimize_bond_length true in particular) may fix it. Adding a constraint to close the ring is an option, but it may be best minimising the whole RNA double helix even at the cost of stuff moving too much.

In your picture is the RNA “blown up”? 

I should have said: relax.minimize_bond_angles(True) and relax.minimize_bond_length(True) result in the model “blowing up” in dihedral space —I see you have cartesian commented out and replaced with dualspace. If the RNA is blown up, then that is what happened —sorry.

If this does nothing and in the original it was not blown up and your backbone ring has indeed an unclosed CUT_BOND pair (O4′ + C1′) then constraining this distance may work, (cf. https://www.rosettacommons.org/node/10951 on adding a constrain without making a constraint file and don’t forget the weight on the scorefunction).