Prediction of partition coefficients in ternary liquid-liquid systems as a function of the phase system composition by factor analysis for

The measured partition data of 26 steroids for six different compositions of ternary liquid-liquid systems consisting of 2,2,4-trimethylpentane, ethanol and water are used in calculations of partition data for new phase system compositions. It was possible to reproduce the original data and to verify new experimental data by this calculation with high accuracy. Calculations were done in three ways. In the first, non-linear regression of the partition coefficients as a function of the phase system composition was examined. With this classical approach it is possible to calculate the values of the partition coefficients, solute by solute, with good accuracy. In the second approach, principal component factor analysis was used. The phase system-specific factors for further phase system compositions were calculated with a non-linear regression of these factors on the phase system composition. In combination with the solute-specific factors, the values of the partition coefficients for all solutes in any phase system composition could be predicted with excellent accuracy. In the third approach, target transformation factor analysis was applied. A relationship between the partition coefficient and the phase system composition was established on the basis of either the phase system composition itself or a simple function of it. This method also gave very accurate predictions of known and unknown partition coefficients.