机器学习辅助乙硫醇高效吸收溶剂分子设计
Machine learning-assisted solvent molecule design for efficient absorption of ethanethiol
陈宇翔 1刘传磊 1龚子君 2赵起越 1郭冠初 1姜豪 1孙辉 3沈本贤1
作者信息
- 1. 华东理工大学石油加工研究所,上海 200237
- 2. 中国科学院过程工程研究所,北京 100190
- 3. 华东理工大学石油加工研究所,上海 200237;华东理工大学绿色能源化工国际联合研究中心,上海 200237;新疆大学石油天然气精细化工教育部重点实验室,新疆 乌鲁木齐 830046
- 折叠
摘要
针对传统胺洗脱硫工艺中有机硫脱除效率低,溶剂开发周期长、成本高等问题,利用7种机器学习算法建立了乙硫醇溶解度的定量构效关系模型,运用SHAP方法阐释了乙硫醇的吸收机理,对备选分子库进行了虚拟筛选,识别出高效吸收脱除乙硫醇的溶剂.基于COSMO-RS模型计算了14732种溶剂的乙硫醇溶解度,这些分子覆盖了广泛的化学空间;XGBoost算法在预测乙硫醇溶解度方面表现最佳,该算法的R2test为0.66,RMSE为1.22,MAE为0.84;分子结构的复杂程度、共价键分布、电荷分布是影响乙硫醇溶解能力的关键因素;确定了4种候选溶剂:3-乙氧基丙胺、3-二乙胺基丙胺、1,4-二甲基哌嗪和3-丁氧基丙胺;平衡溶解度测定实验的结果表明3-丁氧基丙胺的乙硫醇吸收性能最优,亨利常数为37.34 kPa.
Abstract
To solve the problems of low organic sulfur removal efficiency,long solvent development cycle and high cost in the traditional amine elution desulfurization process,the quantitative structure-activity relationship(QSPR)model for ethanethiol solubility was established by using seven machine learning algorithms.Besides,the absorption mechanism of ethanethiol was elucidated by using the SHapley Additive exPlanations(SHAP)method and the virtual screening for candidate molecules was conducted to identify efficient solvents for the absorption removal of ethanethiol.Molar solubilities of ethanethiol in 14732 solvents,which cover a wide range of chemical space,were calculated by using the conductor-like screening model for real solvents(COSMO-RS).XGBoost was identified as the optimal algorithm for predicting the molar solubility of ethanethiol,having R2test of 0.66,RMSE of 1.22,and MAE of 0.84.The complexity of molecular structure,covalent bonding,and electron distribution in molecules were identified as the key factors for the molar solubility of ethanethiol.Four solvents,including 3-ethoxypropylamine,3-diethylaminopropylamine,1,4-dimethylpiperazine,and 3-butoxypropylamine were identified as potential solvents.The results of the equilibrium solubility determination experiments show that 3-butoxypropylamine has the best ethanethiol dissolution with Henry's law constant of 37.34 kPa.
关键词
分子设计/机器学习/溶解度/吸收Key words
molecule design/machine learning/solubility/absorption引用本文复制引用
基金项目
国家自然科学基金项目(21878097)
国家自然科学基金项目(22178109)
上海市自然科学基金项目(21ZR1417700)
出版年
2024
国家科技期刊平台
NSTL
万方数据