We present a new concept of free energy calculations of chemical reactions by means of extended sampling molecular dynamics simulations. Biasing potentials are applied on partial atomic charges, which may be combined with atomic coordinates either in a single collective variable or in multi-dimensional biasing simulations. The necessary additional gradients are obtained by solving coupled-perturbed equations within the approximative density-functional tight-binding method. The new computational scheme was implemented in a combination of Gromacs and Plumed. As a prospective application, proton-coupled electron transfer in a model molecular system is studied. Two collective variables are introduced naturally, one for the proton transfer and the other for the electron transfer. The results are in qualitative agreement with the extended free simulations performed for reference. Free energy minima as well as the mechanism of the process are identified correctly, while the topology of the transition region and the height of the energy barrier are only reproduced qualitatively. The application also illustrates possible difficulties with the new methodology. These may be inefficient sampling of spatial coordinates when atomic charges are biased exclusively and a decreased stability of the simulations. Still, the new approach represents a viable alternative for free energy calculations of a certain class of chemical reactions, for instance a proton-coupled electron transfer in proteins.