To clarify: you have 6 chains in the complex, and you’re docking EF to ABCD?
Which chains are moving 500 angstroms? EF, ABCD, or some combination?
I can think of two possible culprits. (There’s more than I can think of for sure).
One possibility is that the centroid perturbation phase is producing a bad clash, which the minimizer “fixes” by moving the chains apart. The huge step size (500 angstroms) may be due to it being a huge clash or overly granular minimizer settings.
Another is that the code responsible for interpreting which chains are in which rigid body has gone haywire, and is trying to move your whole complex (say, to the origin) but instead accidentally only moves part of it. It could also be separating them deliberately for dG_bind calculations but they somehow aren’t put back together?