Machine learning-assisted high-throughput virtual screening of novel energetic materials.

contextHigh-energy density materials (HEDMs) are indispensable for defense and a wide range of industrial applications. A core challenge in their development is achieving an optimal balance between high energy performance and low mechanical sensitivity. Traditional design approaches suffer from low efficiency, high cost, and potential risks. In this work, we propose an integrated strategy combining density functional theory (DFT) and machine learning (ML) to address these issues. Using 4-(1,5-diamino-1H-1,2,4-triazol-3-yl)-3-(4,5-diamino-1H-1,2,4-triazol-3-yl)-1H-pyrazole as the parent scaffold, we constructed a molecular library of over 10,000 derivatives. A total of 439 representative molecules were selected for property calculation, and five ML algorithms were evaluated. The XGBoost model showed the best predictive performance, with an average test set R

methodsAll quantum chemical calculations were performed at the B3LYP/6-31+G(d,p) level using Gaussian 16. Ten key properties were obtained, including heat of detonation (Q), detonation velocity (D), detonation pressure (P), impact sensitivity (H₅₀), gaseous and solid enthalpies of formation (HOFg, HOFs), enthalpy of sublimation (ΔHsub), density (ρ), oxygen balance (OB₁₀₀), and nitrogen content (N). Five machine learning algorithms-Random Forest (RF), XGBoost, K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Decision Tree (DT)-were assessed using five-fold cross-validation. The optimized XGBoost model was used for high-throughput virtual screening of the entire molecular library, and all top candidates were further validated using the same DFT approach.