机器学习辅助的[5,6]稠环含能化合物高通量设计  

Machine Learning Assisted High-throughput Design of[5,6]Fused Ring Energetic Compounds

作　　者：潘林虎 王睿辉 樊明仁 宋思维 王毅 张庆华 PAN Lin-hu;WANG Rui-hui;FAN Ming-ren;SONG Si-wei;WANG Yi;ZHANG Qing-hua(School of Astronautics,Northwestern Polytechnical University,Xi'an 710072,China)

机构地区：[1]西北工业大学航天学院,陕西西安710072

出　　处：《含能材料》2024年第6期573-583,共11页Chinese Journal of Energetic Materials

基　　金：国家自然科学基金(22205218,22075259,22175157)。

摘　　要：与经验和计算指导的研发模式相比,机器学习辅助的含能分子高通量虚拟筛选技术,在分子设计效率及构效关系定量分析方面都展现出明显优势。鉴于富氮稠环含能化合物较好的能量-稳定平衡特性,研究利用机器学习辅助的高通量虚拟技术对[5,6]富氮稠环类含能分子的化学空间进行了探索研究,基于[5,6]全碳骨架,通过组合枚举和芳香性筛选得到142689个[5,6]稠环类化合物,同时采用核岭回归算法建立并优化了6个含能分子性能预测模型(密度,分解温度,爆速,爆压,撞感和生成焓),分析了稠环上的氮氧原子以及分子上官能团对含能化合物性能的影响。结果发现,所生成稠环化合物的构效关系与含能化合物能量与稳定性相关性的一般规律相符,验证了模型的合理性。以爆速和分解温度作为能量和热稳定性的标准,研究进而筛选获得了5个综合性质较为突出的分子,利用DFT等量子化学计算的结果与本研究模型预测结果符合良好,进一步验证了预测模型的精度。Compared with the research and development model guided by experience and calculations,machine learning-assisted high-throughput virtual screening technology for energetic molecules has shown obvious advantages in terms of molecular design efficiency and quantitative analysis of structure-activity relationships.In view of the fact that nitrogen-rich fused ring energetic compounds usually show better energy-stable balance properties,this study uses machine learning-assisted high-throughput virtual technology to conduct chemical space exploration of[5,6]nitrogen-rich fused ring energetic molecules.Based on the[5,6]all-carbon skeleton,this study obtained 142,689[5,6]fused ring compounds through combined enumeration and aromatic screening.At the same time,a machine learning algorithm was used to establish and optimize an energetic molecular property prediction model(including density,decomposition temperature,detonation velocity,detonation pressure,impact sensitivity and enthalpy of formation).The effects of nitrogen and oxygen atoms on the fused ring and functional groups on the molecule on the performance of energetic compounds were analyzed.The research results show that the structure-activity relationship of the generated fused ring compounds is consistent with the general correlation between energy and stability of energetic compounds,verifying the rationality of the prediction model.Taking detonation velocity and decomposition temperature as the criteria for energy and thermal stability,five molecules with outstanding comprehensive properties were screened,and the quantum chemical calculation results were in good agreement with the machine learning prediction results,which further verified the accuracy of the prediction model.

关 键 词：机器学习 高通量筛选 核岭回归 分子设计 [5 6]稠环含能化合物 

分 类 号：TJ55[兵器科学与技术—军事化学与烟火技术]