FluxDisco: Symbolic Regression for Stoichiometric Dynamical Systems via Monte Carlo Graph Search

Recovering governing differential equations from noisy measurements often yields mathematical formulations that contradict fundamental conservation laws. FluxDisco tackles this issue in stoichiometric ODE systems by embedding known reaction stoichiometries directly into the hypothesis space. By adapting Monte Carlo Graph Search for joint flux discovery, the method maintains physical adherence while successfully identifying interpretable dynamical equations across physical and biological benchmarks.