Prediction of human volume of distribution values for drugs using linear and nonlinear quantitative structure pharmacokinetic relationship models

Abstract

In the present study the volume of distribution values in humans of 121 drugs was estimated using quantitative structure pharmacokinetic relationship analysis. The multiple linear regression (MLR) method and nonlinear artificial neural network (ANN) and support vector machines (SVM) were employed for modeling. The theoretically calculated molecular descriptors were used for modeling and best set of descriptors selected by correlation based feature selection (CFS) method. The performance and predictive capability of linear method was investigated and compared with nonlinear method. The ANN gave better model with an average fold error of 1.66. The test set prediction accuracy shows human volume of distribution values could be predicted, on average, within 2-fold of the actual value.