Prediction of non-ideal behavior of polarity/polarizability scales of solvent mixtures by integration of a novel COSMO-RS molecular descriptor and neural networks

Abstract

A new COSMO-RS descriptor (S_σ-profile) has been used in quantitative structure–property relationship (QSPR) studies based on neural networks (NN) for the prediction of polarity/polarizability scales of pure solvents and mixtures. S_σ-profile is a two-dimensional quantum chemical parameter which quantifies the polar electronic charge on the polarity (σ) scale. Firstly, radial base neural networks (RBNN) are successfully optimized for the prediction of polarizability (SP) and polarity/polarizability (SPP) scales of pure solvents using the S_σ-profile of individual molecules. Subsequently, based on the additive character of the S_σ-profile parameter, we propose to simulate the solvents mixture by the estimation of S^Mixture_σ-profile descriptor, defined as the weighted mean of S_σ-profile values of the components. Then, the SPP parameters for binary and ternary mixtures are accurately predicted using the S^Mixture_σ-profile values into the RBNN model previously developed for pure solvents. As result, we obtain a unique neural network tool to simulate, with similar reliability, the polarity/polarizability of a wide variety of pure organic solvents as well as binary and ternary mixtures which exhibit significant deviations from ideality.