Nonlinear calibration using projection pursuit regression: application to an array of ion-selective electrodes

Abstract

Abstract : Much effort has been expended in the field of analytical chemistry toward the development of selective sensors. The ultimate goal of this area of research is to build sensors that respond to only one analyte while ignoring all other analytes (interferents) that are present in the samples. Perhaps the most common example of the result of this effort is the development of ion selective electrodes (ISE) for the determination of ion concentrations in solutions. While some ISEs are relatively selective for the desired ions, all suffer from some degree of non-specificity. Unfortunately, in the field of sensor development this is a common occurrence. Another approach to solving the problem of interferents is to use multiple non-selective sensors and employ multivariate mathematics to perform the calibration and prediction. This was the approach taken in two recent papers where arrays of ISEs were used to quantify mixtures of analytes. Analyte quantitation was achieved using either linear or non-linear regression techniques to model the response of the electrodes to the concentration of analytes in mixture samples. In both papers, the sensor responses were assumed to obey the relationship found in the set of extended Nernst equations. A relatively new method of regression analysis called projection pursuit regression is introduced. This method performs linear and nonlinear regression and differs from the more classical methods in that it is able to determine the form of the model as well as the model parameters.