Complete characterization for the fit-preserving data refinement of mass-action reaction networks

•We study the fit-preserving data refinement of mass-action reaction networks.•Linear constraints on refined rate constants are sufficient for fit-preservation.•The constraints are necessary for initial models with identifiable rate constants.•Fit-preserving refinements can have qualitatively distinct subspecies dynamics.

The data refinement of reaction-based models consists in substituting species from the original model with several subspecies in the refined one. Fit-preserving refinement, where the goal is to capture the same species dynamics as the original model, helps reduce the computational cost of model fitting by reusing previously fit rate constants. In this paper we give a complete characterization of fit-preserving refinement, as necessary and sufficient linear constraints on the reaction rate constants. Our result is applicable for mass-action reaction networks with uniquely identifiable rate constants. We demonstrate our result on the well-known Brusselator model.