Prediction of IC50 of 2,5-diaminobenzophenone organic derivatives using informatics-aided genetic algorithm

Abstract

In the present paper, informatics-aided quantitative structure activity relationship (QSAR) models using genetic algorithm-partial least square (GA-PLS), genetic algorithm-Kernel partial least square (KPLS), and Levenberg-Marquardt artificial neural network (LM ANN) approach were constructed to access the antimalarial activity (pIC50) of 2,5-diaminobenzophenone derivatives. Comparison of errors and correlation coefficients obtained by the models it was shown that the LM ANN approach works with a high correlation coefficient and low prediction error. This model was applied to the prediction of pIC50 values of 2,5-diaminobenzophenone derivatives. Applying the extended model to a dataset of 20 compounds demonstrate the reliability and accuracy of the model. Comparing three models revealed the superiority of the L-M ANN to predict the pIC50 of 2,5-diaminobenzophenones derivatives.