Prediction of antileukemia activity of berbamine derivatives by genetic algorithm–multiple linear regression

Abstract

A quantitative structure–activity relationship study was performed on a data set of 32 berbamine derivatives that possess antileukemia activity. Semiempirical quantum chemical calculation (AM1 method) was used to find the optimum 3D geometries of the studied molecules. A suitable set of molecular descriptors was calculated and genetic algorithm–multiple linear regression was employed to select the descriptors that resulted in the models with the best fit to the data. A multiple linear regression model with five selected descriptors was obtained. The values of statistical measures such as R 2, Q 2, and F obtained for the training set were within acceptable ranges, so this relationship was applied to the test set. The predictive ability of the model was found to be satisfactory, and it can therefore be used to design similar groups of compounds. Also, results suggest that the charge, electronegativity, and atomic van der Waals volumes of the molecules are the main independent factors that contribute to the antileukemia activity of berbamine derivatives.