Prediction of Explosive Performance and Safety Parameters Using Derivatized SMILES Data

A very large database of about 500 explosives and 135 descriptors were examined to develop a quick model of predicting explosive performance and impact sensitivity. In this model, only data that can be simply calculated from molecular structural information is considered, such as bond environments, molecular formula, functional groups, and oxygen balance, among others. Using random forests, the number of descriptors was reduced to 18, to describe impact sensitivity among all functional group types with similar accuracy to previously reported models (R2= 0.788 and RSME= 0.312). In addition to impact sensitivity, the random forest model accurately predicts important, high explosive parameters, density, heat of formation and heat of explosion. Utilizing the predicted densities and heats of formation, the quick model allows for rapid estimates at detonation velocity, detonation overpressure and impact sensitivity with only a drawn structure.