Data-Driven Combinatorial Design of Highly Energetic Materials

Abstract

In this Account, we present a comprehensive overview of recent advancements in applying data-driven combinatorial design for developing novel high-energy-density materials. Initially, we outline the progress in energetic materials (EMs) development within the framework of the four scientific paradigms, with particular emphasis on the opportunities afforded by the evolution of computer and data science, which has propelled the theoretical design of EMs into a new era of data-driven development. We then discuss the structural features of typical EMs such as TNT, RDX, HMX, and CL-20, namely, a “scaffolds + functional groups” characteristic, underscoring the efficacy of the combinatorial design approach in constructing novel EMs. It has been discerned that those modifications to the scaffolds are the primary driving force behind the enhancement of EMs’ properties.

Subsequently, we introduce three distinct data-driven design strategies for EMs, each with a different approach to scaffold construction. These strategies are as follows: (1) the known scaffold strategy to identify fused cyclic scaffolds containing oxazole or oxadiazole structures from other fields via database screening and employ a high-throughput combinatorial approach with functional groups to design oxazole (and oxadiazole)-based fused cyclic EMs; (2) the semiknown scaffold strategy to construct semiknown scaffolds by integrating known scaffolds and realize the design of bridged cyclic EMs through a high-throughput combination of functional groups; (3) the unknown scaffold strategy to build caged structural models for quantitative characterization, high-throughput screening caged scaffolds from the database, construct unknown caged scaffolds by substituting atoms or substructures, and combine functional groups to design zero oxygen balance caged EMs. Employing the proposed strategies, the design capacity for EMs reaches an impressive scale of 10⁷ molecules, significantly increasing the probability of obtaining high-performance EMs. Furthermore, the incorporation of property assessment models based on machine learning and density functional theory has achieved a balance between computational accuracy and computational speed. Statistical analysis of the virtual screening has revealed the advantages of bicyclic tri- and tetrasubstituted position scaffolds in the construction of high-energy and easily synthesizable fused cyclic EMs. Additionally, the proposed strategies have been successfully applied to design multifunctional modular energetic materials, resulting in the successful synthesis of three target compounds, validating the effectiveness of data-driven combinatorial design approaches.

Lastly, we discuss the current state of high-throughput combinatorial design and, in light of the multifaceted criteria required for the design of EMs, explore the feasibility of multiobjective optimization methods such as Pareto optimization. Moreover, we envision the application of generative models in the subsequent design and development of EMs. We anticipate that this Account will provide valuable insights into the theoretical design of EMs, and we envision the integration of new technologies and methodologies that could play an increasingly significant role in the future discovery of EMs.