Biomolecules are inherently flexible and their dynamics affects ligand binding. To investigate the internal dynamics of proteins and nucleic acids, we apply all-atom molecular dynamics simulations. However, atomistic level of detail makes simulations too computationally demanding to describe folding and global motions so reduced representations of molecules are often applied. These so-called coarse-grained models are sufficient to capture global collective motions on biologically relevant spatial and temporal scales. However, due to reduction of the degrees of freedom, coarse-grained models require parameterizations of the potential energy function (force field). Moreover, coarse-grained force field parameters are typically not transferable between different molecules and problems. I will present our efforts to design an automatic parameterization procedure to obtain force fields for reduced models of biomolecules. The procedure for the optimization of potential energy parameters is based on metaheuristic methods. I will also show examples of applications to dynamics of proteins and nucleic acids.