Since the protein is rigid in most of the docking softwares, I guess this is normal when you start MD due to the potential energy change. You can apply a constraint on ligand-protein COM distance and then you can slowly remove it.

Assuming this is not an artifact of visualization, reconsider equilibration MD protocol, re-check the partial atomic charges of the ligand, find out what triggers the dissociation of the complex, and try several MD runs starting with different initial velocities. There is also the possibility that the binding site or binding mode is not correct, in which case reconsider the docking hypothesis.