Biomolecular soft matter forms the nano-machinery of life. Its main players, i.e., proteins, DNA and RNA, display a complex pattern of self-organization that emerges from first principles of Physics. To understand these phenomena and the associated function of biomolecules, a microscopic description of their multiscale motions is essential. Molecular dynamics simulations provide a powerful approach to predict such dynamics directly from atomistic interactions. However, the necessity of integrating atomistic equations of motion on a femtosecond timescale contrast sharply with biologically relevant timescales of milliseconds to minutes. In this presentation, I will describe new methods to coarse-grain dynamics and accelerate simulations with the help of principles from non-equilibrium statistical mechanics. Using Jarzynski’s equality, we derive a friction term that allows for an on-the-fly calculated dissipation correction. The application to realistic soft matter systems requires further theoretical developments such as a non-equilibrium extension of principal component analysis and machine learning of pathway separation. Langevin equation simulations in combination with temperature rescaling allows to ac-cess molecular dynamics on real timescales up to minutes with affordable computational cost. Furthermore, methods developed here are of interest for the prediction of velocity-dependent properties, such as friction in non-Newtonian liquids.