Testing of model structures in identification of metabolic systems

In this paper some criteria for testing model structures are used for choosing the best among a set of rival dynamical models of two metabolic processes. The considered criteria are based on the residual sum of squares, the final prediction error, the Akaike information criterion and the inverse of the information matrix. Linear time invariant structured models and input-output descriptions are proposed for the two considered metabolic processes, riphamicin and bilirubin kinetics, and maximum likelihood parameter estimates are obtained from input-output test data. A combined use of the Akaike criterion (or final prediction error) and of the inverse of the information matrix proved to be a very reliable tool in the metabolic model structure determination.