Determination of the empirical solvent polarity parameter ET(30) by multivariate image analysis

Abstract

The solvent polarity ET(30) scale has found wide-spread applications in studying chemical processes in solvents. This parameter is usually measured by vis spectrophotometric measurements of the long-wavelength intramolecular charge-transfer (CT) absorption band of Reichardt's pyridinium-N-phenolate betaine dye, e.g. the ET(30) dye, dissolved in the solvent or solvent mixture of interest. Recent advances in colorimetric measurements based on digital photo-capturing devices suggest these methods as a simple, cheap and fast alternative to spectrophotometric measurements in some analytical applications. In this work, we studied the feasibility of colorimetric measurements coupled with multivariate data analysis to determine the empirical solvent polarity parameter ET(30). The picture of the ET(30) dye dissolved in different solvents was captured by a digital camera and then color values in the RGB space were analyzed by the principal component analysis (PCA) method. PCA scores of the unfolded image were then used as input of multiple linear regression and an artificial neural network model to predict the ET(30) parameter. The ANN models were optimized to gain a model of lower prediction ability utilizing a cross-validation test. Then, this was used to predict ET(30) values for an external solvent test set. The generated model could explain and predict 99% of the variances in the polarity data and can predict ET(30) values with a root mean square error of 2.25 kcal mol−1 (in the ET(30) scale). The results suggest colorimetric measurements as a useful and practical alternative to the vis spectrophotometric measurements for determination of solvent polarity parameters derived from solvatochromic betaine dyes.