Predicting the Melting Point of Energetic Molecules Using a Learnable Graph Neural Fingerprint Model

Abstract

Melting point prediction for organic molecules has drawn widespread attention from both academic and industrial communities. In this work, a learnable graph neural fingerprint (GNF) was employed to develop a melting point prediction model using a dataset of over 90,000 organic molecules. The GNF model exhibited a significant advantage, with a mean absolute error (MAE) of 25.0 K, when compared to other featurization methods. Furthermore, by integrating prior knowledge through a customized descriptor set (i.e., CDS) into GNF, the accuracy of the resulting model, GNF_CDS, improved to 24.7 K, surpassing the performance of previously reported models for a wide range of structurally diverse organic compounds. Moreover, the generalizability of the GNF_CDS model was significantly improved with a decreased MAE of 17 K for an independent dataset containing meltcastable energetic molecules. This work clearly demonstrates that prior knowledge is still beneficial for modeling molecular properties despite the powerful learning capability of graph neural networks, especially in specific fields where chemical data are lacking.

Predicting the enthalpy of formation of energetic molecules via conventional machine learning and GNN

“…This suggests that handcrafted descriptor based on prior knowledge is more rational and concise in capturing the relationship between molecular features and properties. This was also proven in the works of Song et al , 30 Tian et al , 22 and Nguyen et al 26 In the original work of Casey et al , 27 they reported a prediction error on the same data set as 8.07 and 15.0 kcal mol −1 on the training and test set, respectively. They combined a 3D CNN model with charge density and electrostatic potential as the descriptor and the descriptor was calculated using density functional theory at the B3LYP/6-31G** level.…”

“…In addition, there are a variety of ML methods (both conventional and emerging) that directly fit target data to predict chemical properties. Conventional ML techniques, including random forest (RF) model, support vector machine (SVM) model, and multi-layer perceptron (MLP) model, have demonstrated successful applications in predicting properties such as EOF, 22–24 density, 25,26 detonation velocity, 27,28 sensitivity, 29 melting point, 30,31 solubility, 32 entropy and heat capacity. 33,34 These conventional ML models are typically based on descriptors, which define the molecules prior to fitting an ML model for prediction.…”

Insight into melting point differences of dinitroimidazoles and dinitropyrazoles from the perspective of intermolecular interactions

“…Consequently, chemists are making concerted efforts to investigate diverse innovative approaches for the prediction of melting points. 1–9 Currently, molecular dynamics, are believed to be the most powerful method for predicting melting points. 10–12…”