Quantitative structure–property relationship study of β-cyclodextrin complexation free energies of organic compounds

Abstract

A quantitative structure–property relationship (QSPR) study was performed between β-cyclodextrin complexation free energies and descriptors representing the molecular structures of organic guest compounds. The entire set of 218 compounds was divided into a training set of 160 compounds and a test set of 58 compounds by DUPLEX algorithm. Multiple linear regression (MLR) analysis was used to select the best subset of descriptors and to build linear models; while nonlinear models were developed with artificial neural network (ANN). The obtained models with seven descriptors involved show good predictive power for the test set: a squared correlation coefficient (r2) of 0.833 and mean absolute error (MAE) of 1.911 was achieved by the MLR model; while the ANN model performed better than the MLR model, with r2 of 0.957 and MAE of 0.925 for the test set. In addition, the applicability domain of the models was analyzed based on the Williams plot.