Robustly interpretable melting-point prediction for structurally diverse nitro-containing compounds

Accurate melting point prediction for nitro-containing flexible organic molecules is critical for optimizing their performance in pharmaceuticals, energetic materials, and chemical processing. Using nine representative nitro compounds tailored to explore intermolecular interactions and the effect of vibrational entropy effects on the melting point of nitro-containing flexible compounds, we developed a dual descriptor framework that integrates enthalpy descriptors (from crystal structure) and entropy descriptors (from ab initial molecular dynamics, AIMD). Through comparison between the linear multivariate regression (MLR) model and the nonlinear KNN model, combined with SHAP analysis, the KNN model, which integrates intermolecular interactions (BEC_max1-3, BEC_total), molecular van der Waals volume (VW), and molecular vibrational features (RMSD_ave, E_b, and ΔE)—demonstrated high accuracy (training set: R2 = 0.83, MAE = 12.48 °C; test set: R2 = 0.79, MAE = 15.29 °C) and interpretability. Mechanistic analysis indicates that the melting process is dominated by the collaborative effect of enthalpy and entropy. This framework establishes a structure-property linkage for rational design of new nitro-containing flexible organic compounds.