Prediction of the net heat of combustion of organic compounds based on atom-type electrotopological state indices

An accurate quantitative structure-property relationship (QSPR) model, based on the atom-type electrotopological state (E-state) indices and artificial neural network (ANN) technique, for prediction of standard net heat of combustion (ΔH_c^o) was developed. An extended set of 49 atom-type electrotopological state (E-state) indices that combined together both electronic and topological characteristics of the analyzed molecules were used as molecular structure descriptors for a diverse set of 1496 organic compounds. Both multilinear regression (MLR) and artificial neural network (ANN) were employed in the modeling. The ANN model with the final optimum network architecture of [49-35-1] gave a significant better performance than the MLR model. The squared correlation coefficient R² for the ANN model was R² = 0.991 for the training set of 1196 compounds. For the test set of 300 compounds, the corresponding statistics was R² = 0.992. The results of this study showed that it would be successful to predict ΔH_c^o by using the easily calculated atom-type E-state indices, which can provide one more way for predicting the ΔH_c^o of organic compounds for engineering based on only their molecular structures.