Machine-learning-driven QSPR models for energetic molecules: A review on safety and energetic properties prediction

Abstract

The performance prediction and rational design of energetic molecules (EMs) remain central challenges in their development. Traditional experimental methods are constrained by prohibitively high costs and inherent safety risks, highlighting the urgent requirement for efficient computational alternatives. The quantitative structure-property relationship (QSPR) framework allows quantitative mapping between the molecular structures and macroscopic properties of EMs. When integrated with machine learning (ML), a methodology uniquely suited for deciphering complex structure-property correlations, the resulting combination facilitates the accurate prediction of EMs behavior. This review critically evaluates ML-driven QSPR models for predicting energetic properties (e.g., enthalpy of formation, sublimation enthalpy, density, and detonation velocity) and safety characteristics (e.g., impact sensitivity and thermal stability) of EMs. The key technical aspects are systematically analyzed, including molecular descriptor optimization, ML algorithm selection, and model validation strategies. Persistent challenges, such as data scarcity and model generalizability, are thoroughly discussed. Finally, future directions are outlined for the development of high-precision QSPR models through a synergistic data algorithm-application paradigm, emphasizing multi-objective optimization and interpretable ML frameworks.