Accelerated discovery of high-density pyrazole-based energetic materials using machine learning and density functional theory

Pyrazole-based energetic materials are heterocyclic aromatic compounds characterized by unique structural configuration that promotes high energy content per unit mass and crystal density. These desirable properties render them particularly suitable candidates for energetic materials, with potential applications in explosives and propellants. In this study we developed an efficient strategy that integrate data-driven approach with density functional theory to design novel pyrazole-based energetic materials. Using genetic function approximation algorithm, pertinent molecular descriptors were identified and used to build robust Quantitative Structure Property Relationship (QSPR) models for predicting crystalline density of energetic materials. The performance of four machine learning algorithms including: multilinear regression, artificial neural network, support vector machines, and random forest algorithms were evaluated. The results indicate the best predictive performance was afforded by random forest algorithm with Pearson’s correlation coefficient (RTR), Cross-validation coefficient (QCV) and External validation coefficient (QEX) values of 0.9273, 0.7294 and 0.7184 respectively. Using the compound with highest crystalline density in the dataset, novel energetic materials were designed. The crystalline density of the designed compounds was predicted using ML and DFT approach. The values predicted using high level DFT/B3PW91/6-311 g level of theory, ranging from 2.0389–2.3164gcm−3, were found to be closer to experimental values. Electronic structure and quantitative electrostatic potential investigations indicated the designed compounds possessed favorable properties for high performing energetic materials. This integrated approach to in-silico design of energetic materials could accelerate the discovery of high performing energetic materials.