Search

The combination of unfolded partial least-squares U-PLS with residual bilinearization RBL has not been properly exploited to process experimental second-order spectroscopic information, although it is able to achieve the important second-order advantage. Among other desirable properties, the technique can handle incomplete calibration information, i.e., when only certain analyte concentrations are known in the training set. It can also cope with analyte spectral changes from sample to sample, due to its latent variable structure. In this work, U-PLSRBL has been successfully applied to experimental fluorescence excitation–emission matrix data aimed at the quantitation of analytes in complex samples: these were the antibiotic tetracycline and the anti-inflammatory salicylate, in both cases in the presence of human serum, where significant analyte–background interactions occur. The interactions of the analyte with the serum proteins modify their spectral fluorescence properties, making it necessary to employ training sets of samples where the biological background is present, possibly causing analyte spectral changes from sample to sample. The predictive ability of the studied model has been compared with that of parallel factor analysis PARAFAC, as regards test samples containing different sera, and also other pharmaceuticals which could act as potential interferents.