Coarse-grained model optimization

With ThermoDiff, CG models can be improved to match either experimental data (in a similar way to AA force field reparametrization) or AA-derived quantities, such as ensemble averages or profiles of any observables, as well as free energies.

A typical procedure aimed at improving the parameters of a particular CG model would be to perform a set of all-atom enhanced sampling simulations (e.g. umbrella sampling, or replica-exchange metadynamics coupled to an equilibrium ensemble) and then ask ThermoDiff to reproduce these quantities on the CG level; it is particularly reasonable to try to optimize the NBFIX (i.e. custom combination rules) components, as these allow to fine-tune the interactions between specific beads without affecting global properties of the molecular model. If one however aims to optimize global properties, the optimization of \(\sigma\), \(\varepsilon\) and atomic charges can also be accomplished with ThermoDiff.