Scholarlink

<p>The development of high energy density materials relies on the discovery of molecules with low gas-phase enthalpy of formation (<i>ΔH</i><sub><i>f</i></sub><i>°</i>). In this study, a machine learning (<i>ML</i>) approach using Python has been applied to predict the <i>ΔH</i><sub><i>f</i></sub><i>°</i> of energetic molecules. Various models, including Ridge, Elastic, Lasso, and Historical Gradient Boosting regressors, accurately predicted the enthalpy of formation for 300 energetic molecules, achieving a coefficient of determination (R<sup>2</sup>) of 0.68. SHapley impact analysis reveals that AATSC0pe and AMID_O features exhibit the highest correlations with the predicted enthalpy of formation. The results are further validated by using five fingerprint methods, including Morgan Fingerprints, and demonstrated an increase in model performance with training size. Using the Breaking Retro Synthetically Interesting Chemical Species (<i>BRICS</i>) approach, 1561 new energetic compounds with <i>ΔH</i><sub><i>f</i></sub><i>°</i> as low as -1542 <i>kJ/mol</i> has been designed. Notably, the results show that molecules with a <i>SMILES</i> length of less than 1000 can exhibit low <i>ΔH</i><sub><i>f</i></sub><i>°</i>, expanding the chemical space for high energy density materials design. This work demonstrates the potential of <i>ML</i> and computational chemistry for the rapid discovery of novel energetic molecules with tailored properties.</p> Graphical Abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Designing low gas-phase enthalpy of formation based energetic molecules in python for high energy density materials chemical space

  • 期刊论文

  • 发表日期：

    2025年10月31日

查看全文
  • Shaimaa H. Mallah,
  • Azal S. Waheeb,
  • Afaf M. Kadhum,
  • Hussein A.K. Kyhoiesh,
  • Hassan E. Abd Elsalam,
  • Islam H. El Azab

摘要

The development of high energy density materials relies on the discovery of molecules with low gas-phase enthalpy of formation (ΔHf°). In this study, a machine learning (ML) approach using Python has been applied to predict the ΔHf° of energetic molecules. Various models, including Ridge, Elastic, Lasso, and Historical Gradient Boosting regressors, accurately predicted the enthalpy of formation for 300 energetic molecules, achieving a coefficient of determination (R2) of 0.68. SHapley impact analysis reveals that AATSC0pe and AMID_O features exhibit the highest correlations with the predicted enthalpy of formation. The results are further validated by using five fingerprint methods, including Morgan Fingerprints, and demonstrated an increase in model performance with training size. Using the Breaking Retro Synthetically Interesting Chemical Species (BRICS) approach, 1561 new energetic compounds with ΔHf° as low as -1542 kJ/mol has been designed. Notably, the results show that molecules with a SMILES length of less than 1000 can exhibit low ΔHf°, expanding the chemical space for high energy density materials design. This work demonstrates the potential of ML and computational chemistry for the rapid discovery of novel energetic molecules with tailored properties.

Graphical Abstract