化学
选择性
催化作用
乙烯
吉布斯自由能
过渡金属
特征选择
组合化学
生物系统
计算化学
生化工程
人工智能
热力学
计算机科学
有机化学
工程类
物理
生物
作者
Zhihan Luo,Jiale Peng,Yao Mu,Li Sun,Zhihua Zhu,Zhen Liu
标识
DOI:10.1016/j.jcat.2023.115127
摘要
The implementation of data-driven methods in the domain of transition metal catalyst design has emerged as an undeniable trend. With the Curtin-Hammett principle, the selectivity of ethylene tri-/tetramerization showed a strong correlation with the relative Gibbs free energy (ΔΔG) of the key transition states. By leveraging the machine learning method, the prediction of ethylene tri-/tetramerization can be accomplished through a database training model, thereby expediting the process of new catalysts design. Herein, we constructed a group of practical descriptors that exhibit a close relationship with the Gibbs free energy, proving to be more valuable in the design of new ligands when compared to extracting elusive descriptors from the sophisticated molecular structure. Through high-throughput screening feature selection, we established an XGBoost machine learning model for Cr bisphosphine (Cr-PNP) catalysts, enabling the power of the prediction of selectivity for ethylene tri-/tetramerization. In this work, the descriptors extracted from precatalysts only take into account the influence of the metal center and the ligands, avoiding the complicated and laborious conformational search required for predicting the selectivity of the new ligands, which effectively reduces the computational costs. Descriptor analysis, guided by the feedback from the model, allows us to identify the most influential factors governing the selectivity, which can be regulated to effectively advance the design of new catalysts. The validation performed with new ligands confirms the well-predictive performance of the model, as evidenced by the relatively low mean absolute error (MAE).
科研通智能强力驱动
Strongly Powered by AbleSci AI