Single-sensor voltammetric electronic tongue based on the pH variations and chemometrics preprocessing for the classification of the pharmaceutical samples by the neural networks

The imitation of the taste sensation process in human and animals can be used as an analytical tool for qualitative evaluation and classification of the food, drink and chemical samples. In this manner, the voltammetric Electronic Tongues (ETs) consisting of a set of solid-state electrodes with different electrochemical properties, have been extensively used in the analysis of the various real samples [1]. The IUPAC report defines an electronic tongue as “a multisensor system, which consists of a number of low-selective sensors and uses advanced mathematical procedures for signal processing based on the pattern recognition and/or multivariate data analysis”. Based on this definition, any parameter like pH or ionic strength of the solution, which induce a subtle difference in the voltammetric response of the electrode toward a specific analyte, can be used to fabricate an electronic tongue that uses the obtained responses to create a fingerprint for each of the analytes [2, 3]. As most of the analytes undergo the structural variations due to the change in the pH of the medium, which results in the different voltammetric behaviors, this parameter (pH) can be employed to design a voltammetric electronic tongue for the classification purposes without the need to the expensive and time-consuming electrode preparation and optimization procedures for each of the sensors. In this study a single glassy carbon electrode (GCE) modified with the MWCNTs was used for the classification of 14 different drugs at three concentrations in their pure form and binary mixtures using the Linear Sweep Voltammograms (LSV) recorded at three different pH values and three replications in the ranges 0.2 to 1.2 V. The obtained large data matrix (126×2985) was preprocessed by the first-order derivatization to reduce the overlapping and background interference and Discrete Wavelet Transform (DWT) with the Coiflet3 mother wavelet at five decomposition levels for the dimensionality reduction, which results in an about 96% compression ratio. The obtained approximation coefficients (126×109) together with the raw and first-derivative data matrices were analyzed by both PCA and XY-Fused neural networks as unsupervised and supervised pattern recognition techniques respectively. The results indicate that the XY-Fused neural networks due to their nonlinear characteristics can better classify the analyzed samples using their processed voltammetric responses compared to the PCA as a linear data processing chemometrics technique.