Toward the Identification of Stoichiometric Models for Complex Reaction Mixtures

We present an approach to identify stoichiometric and kinetic models that accurately represent the composition data from batch experiments without a priori knowledge of candidate stoichiometries. Possible reactions between the measured species that fulfill mass balance constraints and pass a target factor analysis (TFA) test are considered. From the qualified reaction candidates, reaction networks are enumerated. Four kinetic models, including reversible and irreversible kinetics, are estimated for each reaction within a candidate network. The most accurate kinetic models for each network are chosen using the Bayesian information criterion. Reaction networks are subsequently ranked according to their regression sum of squares. The kinetic parameters of the most accurate networks are re-estimated in a centralized fashion against the measured concentrations. An F test between the networks' regression sums of squares reveals the most accurate network. Two case studies are analyzed, involving three and six species participating in two and three reactions. The proposed algorithm identifies the actual stoichiometric and kinetic model for both systems.

Automated summary

We present an approach to identify stoichiometric and kinetic models accurately from batch experiments without prior knowledge of candidate stoichiometries. This method considers possible reactions between measured species that fulfill mass balance constraints and pass a target factor analysis test.