Target factor analysis for the identification of stoichiometric models

Abstract

A methodology for identifying stoichiometric models for complex reaction systems is developed. The approach is useful for (1) deriving simple, approximate stoichiometric models for complex systems, or (2) testing possible stoichiometries and thereby investigating reaction pathways. The method of factor analysis is used to determine the number of reactions and derive an observed stoichiometric space from measured composition and possibly thermal data. The validity of proposed target stoichiometries can then be tested for compatibility with this stoichiometric space. The approach is straightforward with noise-free data. Various ways of coping with measurement errors are also presented. A major contribution of this paper consists in an approach for incorporating known stoichiometric information, thereby reducing considerably the effect of measurement noise. The methodology is tested on a simulated model of an industrial system. It is found to be very useful for clarifying the unknown part of the stoichiometric model. Extensions that should facilitate the applicability of the approach to real industrial systems are also discussed.