The rapid evolution of molecule's imaging and observation's techniques has caused a growing interest in studying molecular structures. Naturally, scientists have turned to simulation and 3D modeling in order to better understand biological phenomena. Thus, several 3D modeling systems have emerged. Some of these systems are dedicated to 3D visualization, and others are interested in 3D handling. However, we observed that these systems use classical 3D interaction techniques, frequently used in virtual reality (VR). On the other hand, the biological environment is very complex and binding as well. Thus, to remain faithful to the constraintes of the environment and be closer to natural behavior of molecules, we have tried to propose a 3D manipulation adapted to the domain, a bio-supervised 3D manipulation.