Comparison of the performance of an artificial neural network and multiple linear regression in the prediction of the biological activity of cocaine analogues from molecular descriptors

objective of this investigation was to compare the performance of artificial neural networks against multiple linear regression in predicting the biological activity of cocaine analogues from molecular descriptors. For this purpose, a set of 14 molecular descriptors grouped into quantum chemical descriptors and descriptors of the three-dimensional structure of the molecule were selected and their values were calculated theoretically for 65 cocaine analogue structures, followed by the construction of the artificial neural networks model and multiple linear regression for the prediction of biological activity expressed as affinity (IC50). It was found that the artificial neural networks had an R2 of 0,8651 while the linear multiple regression had an R2 value of 0,039, showing that artificial neural networks perform better than linear multiple regression in the prediction of the biological activity of cocaine analogues from the selected molecular descriptors, and that the effect of these descriptors on the biological activity is non-linear in nature.