Predicting the electric spark sensitivity of nitramines from molecular structures via support vector machine

• Five most relevant molecular descriptors were selected by genetic algorithm.
• The electric spark sensitivity is mainly affected by molecular connectivity and charge distribution.
• The SVM model shows good prediction ability, robustness and generalization through the model validation.
• The present method possesses some superiority in terms of prediction precision, model generalization, as well as time saving.

A new model is constructed to predict the electric spark sensitivity of nitramines. Genetic algorithm was employed to select the optimal subset of descriptors which have significant contribution to the electric spark sensitivity from various calculated molecular structure descriptors. The novel modeling method of support vector machine was then applied to model the possible quantitative relationship between selected descriptors and electric spark sensitivity. The results are satisfactory in terms of prediction capability, robustness, and generalization. The new model was also compared with previous ones. The comparison results indicate the superiority of the present model and reveal that it can be effectively used to predict the electric spark sensitivity of nitramines from the molecular structures alone.