Identification of biochemical reaction networks: an observer based approach

Dynamic models present a fundamental tool in systems biology, but rely on kinetic parameters, such as association and dissociation constants. Their direct estimation from studies on isolated reactions is usually expensive, time-consuming or even infeasible for large models. As a consequence, they must be estimated from indirect measurements, usually in the form of time-series data. We describe an observer-based parameter estimation approach taking the specific structure of biochemical reaction networks into account. Considering reaction kinetics described by polynomial or rational functions, we propose a three step approach. In a first step, the estimation of not directly measured states is decoupled from the estimation of the parameters using a suitable model extension. In a second step, a specially suited nonlinear observer estimates the extended state. Based on the obtained state estimates, the parameter estimates are calculated in a straightforward way in the final step. The applicability of the approach is exemplified considering a simplified model of the circadian rhythm.