Quantitative structure property relations (QSPR) for predicting molar diamagnetic susceptibilities, χm, of inorganic compounds

For predicting the molar diamagnetic susceptibilities of inorganic compounds, a novel connectivity index (m)G based on adjacency matrix of molecular graphs and ionic parameter g(i) was proposed. The g(i) is defined as g(i)= (n(i)(0.5)-0.91)(4).x(i)(0.5)/Z(i)(0.5), where Z(i), n(i), x(i) are the valence, the outer electronic shell primary quantum number, and the electronegativity of atom i respectively. The good QSPR models for the molar diamagnetic susceptibilities can be constructed from (0)G and (1)G by using multivariate linear regression (MLR) method and artificial neural network (NN) method. The correlation coefficient r, standard error, and average absolute deviation of the MLR model and NN model are 0.9868, 5.47 cgs, 4.33 cgs, 0.9885, 5.09 cgs and 4.06 cgs, respectively, for the 144 inorganic compounds. The cross-validation by using the leave-one-out method demonstrates that the MLR model is highly reliable from the point of view of statistics. The average absolute deviations of predicted values of the molar diamagnetic susceptibility of other 62 inorganic compounds (test set) are 4.72 cgs and 4.06 cgs for the MLR model and NN model. The results show that the current method is more effective than literature methods for estimating the molar diamagnetic susceptibility of an inorganic compound. Both MLR and NN methods can provide acceptable models for the prediction of the molar diamagnetic susceptibilities. The NN model for the molar diamagnetic susceptibilities appears more reliable than the MLR model.