Markov V. Quantitative «structure of substance – partition characteristics» relationships in two-phase and nanodisperse systems

Existing models to predict physical-chemical properties of substances were analyzed. Linear solvation energy relationships (LSER) were used to describe mobile phase modificator influence on analyte retention. Model coefficient variance on system composition change well corresponds to variance of physical-chemical properties. Methods better fitted to solving of thesis problem were selected. Genetic algorithm and brute force search method were optimized to select a set of molecular descriptors and construct multiparametric «structure – retention – property relationships». The models were built to predict distribution of organic substance in two-phase and micellar systems, as well as to model chromatographic retention. Models are universal, so they are applicable to systems of different types. The proposed models are not inferior to classical linear solvation energy relationships, but based on non-empirical molecular descriptors. The constructed relationships were validated by predicting of the distribution of carboxylic acids in «water – SDS micellar pseudophase» system, cross-validation, prediction of distribution constants for the independent set of substances. The predicted distribution constants are in good agreement with experimentally measured.