QSPR modeling of octanol/water partition coefficient of antineoplastic agents by balance of correlations

Three different splits into the subtraining set (n = 22), the set of calibration (n = 21), and the test set (n = 12) of 55 antineoplastic agents have been examined. By the correlation balance of SMILES-based optimal descriptors quite satisfactory models for the octanol/water partition coefficient have been obtained on all three splits. The correlation balance is the optimization of a one-variable model with a target function that provides both the maximal values of the correlation coefficient for the subtraining and calibration set and the minimum of the difference between the above-mentioned correlation coefficients. Thus, the calibration set is a preliminary test set.

The determination coefficient (square of the correlation coefficient) between optimal SMILES-based descriptor and octanol/water partition coefficient for an external test set of antineoplastic agents can be improved by the balance of correlations.Figure optionsDownload as PowerPoint slide