Searching for the analogues of 1,1-dinitro-2,2-diamino ethylene (FOX-7) by high-throughput computation and machine learning

Abstract

1,1-dinitro-2,2-diamino ethylene (FOX-7) is typically representative of low sensitivity and high energy compound. In this work, analogues of FOX-7 are screened using a combined method of high-throughput computation (HTC) and machine learning (ML). The molecules are generated with typical unsaturated hydrocarbons backbones and random combination of substituents -H, -NH2 and -NO2, then HTC is performed based on 200 sample molecules. ML models are established based on the HTC results, with detonation parameters predicted using the most accurate model of extreme gradient boosting (XGB). Finally, stability of the filtered high energy molecules are confirmed by quantum chemistry calculations, and besides FOX-7, 8 more energetic molecules with high energy as well as high stability (detonation velocity ≥ 8841.1 m/s, detonation pressure ≥ 34.6 GPa and stability parameter bond dissociation energy ≥ 201.7 kJ/mol) are achieved. This work has shown the efficiency of HTC and ML methods in searching new target molecules.

Keywords

• Energetic materials
• 1,1-dinitro-2,2-diamino ethylene (FOX-7)
• Molecular design
• High-throughput computation
• Machine learning