Introducing nonlinear, multivariate 'Predictor Surfaces' for quantitative modeling of chemical systems with higher-order, coupled predictor variables.

Innovations in chemometrics are required for studies of chemical systems which are governed by nonlinear responses to chemical parameters and/or interdependencies (coupling) among these parameters. Conventional and linear multivariate models have limited use for quantitative and qualitative investigations of such systems because they are based on the assumption that the measured data are simple superpositions of several input parameters. 'Predictor Surfaces' were developed for studies of more chemically complex systems such as biological materials in order to ensure accurate quantitative analyses and proper chemical modeling for in-depth studies of such systems. Predictor Surfaces are based on approximating nonlinear multivariate model functions by multivariate Taylor expansions which inherently introduce the required coupled and higher-order predictor variables. As proof-of-principle for the Predictor Surfaces' capabilities, an application from environmental analytical chemistry was chosen. Microalgae cells are known to sensitively adapt to changes in environmental parameters such as pollution and/or nutrient availability and thus have potential as novel in situ sensors for environmental monitoring. These adaptations of the microalgae cells are reflected in their chemical signatures which were then acquired by means of FT-IR spectroscopy. In this study, the concentrations of three nutrients, namely inorganic carbon and two nitrogen containing ions, were chosen. Biological considerations predict that changes in nutrient availability produce a nonlinear response in the cells' biomass composition; it is also known that microalgae need certain nutrient mixes to thrive. The nonlinear Predictor Surfaces were demonstrated to be more accurate in predicting the values of these nutrients' concentrations than principal component regression. For qualitative chemical studies of biological systems, the Predictor Surfaces themselves are a novel tool as they visualize nonlinearities and more importantly the coupling among predictor variables. Thus, they can serve as a novel tool for studies in bioanalytical chemistry, biology, and ecology.