Acta Pharmacologica Sinica 2008 March; 29 (3): 385-396; doi: 10.1111/j.1745-7254.2008.00746.x

Aim: To develop an artificial neural network model for predicting the resistance index (RI) of taxoids.

Methods: A dataset of 63 experimental data points were compiled from published studies and randomly subdivided into training and external test sets. Electrotopological state (E-state) indices were calculated to characterize molecular structure together with a principle component analysis to reduce the variable space and analyze the relative importance of E-state indices. Back propagation neural network technique was used to build the models. Five-fold cross-validation was performed and 5 models with different compound composition in training and validation sets were built. The independent external test set was used to evaluate the predictive ability of models.

Results: The final model proved to be good with the cross-validation Q²_cv0.62, external testing R² 0.84, and the slope of the regression line through the origin for the testing set at 0.9933.

Conclusion: The quantitative structure-activity relationship model can predict the RI to a relative nicety, which will aid in the development of new anti-multidrug resistance taxoids.

Keywords: artificial neural network model; taxoids; multidrug resistance; resistance index; electrotopological state indices; principle component analysis; quantitative structure-activity relationship