Classical Least Squares Combined with Spectral Interval Selection Using Genetic Algorithm for Prediction of Constituents in Pharmaceutical Solid Dosage Forms from near Infrared Chemical Imaging Data

Abstract

A new algorithm that combines spectral interval selection using genetic algorithm and classical least squares (GA-iCLS) is presented for the prediction of the active pharmaceutical ingredients and excipients in various pharmaceutical solid dosage forms from near infrared chemical imaging data. The algorithm is based on the CLS approach, selecting the best wavenumber intervals in the unfolded hyper-cube of each sample (D), and in pure-compound reference spectra (S), wherein the pixel-to-pixel prediction capability of the compounds, obtained by C = DST(SST)−1, is optimised for the samples. The wavelength intervals were selected (GA optimisation) while minimising the error between the mean concentrations of the ith compound predicted in the pixels and the nominal concentration in the corresponding sample (known a priori). The excluded wavenumber intervals from D (and S), for each sample, were interpreted based on systematic deviations from D = CST + E (CLS approach) due to the scattering effects and/or intermolecular interactions in mixtures of the pure compounds. The comparison of the chemical images generated from the predictions performed using the GA-iCLS algorithms with similar images obtained without spectral interval selection, using direct CLS and multivariate curve resolution–alternating least squares, revealed the potential applicability of the proposed algorithm for analytical purposes for pharmaceuticals using chemical imaging data.