Molecular descriptor-enhanced graph neural network for energetic molecular property prediction

Energetic molecules (EMs) play an important role in both military and civilian applications. Traditionally, determining the physicochemical parameters of EMs requires experimental workload and inherent risks while new-rising machine learning (ML) methods are promising to address this challenge. In this work, we report a molecular descriptor-enhanced graph neural network (MD-enhanced GNN) model to accurately and fast predict three detonation parameters of EMs. This model integrates sequence-based molecular descriptors and structure-based graph vectors, offering a comprehensive framework that does not require custom descriptors. Accordingly, we construct an EMs dataset that includes 18,991 CHNO EMs and compare our model with sole molecular fingerprint/descriptor and GNN methods. It is found that our proposed MD-enhanced GNN integration method achieves superior accuracy with R-2 over 0.93 and a learning speed improvement of over 20% by combining two different complementary features, which highlights the potential of our model in reshaping the landscape of EMs design, promising substantial improvements in both efficiency and effectiveness within this critical field.