I am trying to do an optimization while constraining the distance between two atoms on two separate non-connected fragments, using the open babel python bindings with MMFF94 and conjugate gradients, with and without a preliminary random rotor search. If this distance in the input structure is close (~0.1Å) to the constraint requirement, things turn out fine, but things go badly wrong when it's not close; the molecule basically blows up, and other curiosities occur, such as the CG optimization being both instantaneous and independent of changing the number of steps from 5000 to 12000.
I have attached the python script and two initial structures, one that works and one that fails.
Any help or testimonies of similar experiences would be greatly appreciated, as I am completely out of ideas at this point.
Best regards, Kasper Thofte University of Copenhagen