Author:

Keywords:

chemometrics, interferents, pure component, spectroscopy, NIR, multivariate calibration

Abstract:

A class of multivariate calibration methods called augmented classical least squares (ACLS) has been proposed which combines an explicit linear additive model with the predictive power of inverse models, such as principal component regression (PCR) and partial least squares (PLS). Thanks to its use of the explicit linear additive model, ACLS provides an interesting framework to incorporate different sources of prior information, such as measured pure component spectra, in the model. A measured spectrum can then be decomposed using the following formula: A = CK + CpKp + CiKi + TP + E where CK corresponds to those constituents for which the concentrations C are known, but the pure component spectra K are unknown. The subscripts P and I refer to components (known concentration), respectively interferents (unknown concentration) for which measured pure component spectra are available. Thus, CP is the matrix of known concentration vectors for those components with measured pure spectra KP, and CI is the matrix of unknown concentrations for the interferents with measured pure component signals included in KI. The TP+E contribution corresponds to a PCA model for those interferents that could not be modelled by the other terms in the model. The T’s are the scores, the P contains the loadings and the E matrix can be interpreted as noise after modelling of the significant components. In this study, the predictive power of ACLS models incorporating different amounts of prior information has been compared to that of PCR and PLS using two examples: a designed experiment and one with biological samples. In both cases, the ACLS models showed predictive power comparable to PLS under idealised validation conditions. When a different interferent structure was present in the validation samples the predictive power of the inverse models (PCR, PLS) dramatically decreased, with an increase in RMSEP of a factor 3.5 for the first example and a factor 2 in the second example. The incorporation of prior information in the ACLS framework was found to considerably reduce or even completely remove these dramatic effects, especially when the pure component contributions for the interferents were taken into account.