Pure component property estimation framework using explainable machine learning methods

doi:10.1016/j.cjche.2025.05.011

中国化学工程学报 ›› 2025, Vol. 84 ›› Issue (8): 158-178.DOI: 10.1016/j.cjche.2025.05.011

• Full Length Article • 上一篇下一篇

Pure component property estimation framework using explainable machine learning methods

Jianfeng Jiao¹, Xi Gao^2,3, Jie Li¹

1. Centre for Process Integration, Department of Chemical Engineering, School of Engineering, The University of Manchester, Manchester M13 9PL, UK;
2. School of Electronic and Information Engineering, Tongji University, Shanghai 201804, China;
3. School of Mechanical and Electrical Engineering, Jinggangshan University, Ji'an 343009, China

收稿日期:2025-02-17 修回日期:2025-05-12 接受日期:2025-05-13 出版日期:2025-08-28 发布日期:2025-06-06
通讯作者: Xi Gao,E-mail:9920000016@jgsu.edu.cn;Jie Li,E-mail:jie.li-2@manchester.ac.uk
基金资助:
The authors gratefully acknowledge the financial support from China Scholarship Council (CSC) (202406440073).

Pure component property estimation framework using explainable machine learning methods

Jianfeng Jiao¹, Xi Gao^2,3, Jie Li¹

1. Centre for Process Integration, Department of Chemical Engineering, School of Engineering, The University of Manchester, Manchester M13 9PL, UK;
2. School of Electronic and Information Engineering, Tongji University, Shanghai 201804, China;
3. School of Mechanical and Electrical Engineering, Jinggangshan University, Ji'an 343009, China

Received:2025-02-17 Revised:2025-05-12 Accepted:2025-05-13 Online:2025-08-28 Published:2025-06-06
Contact: Xi Gao,E-mail:9920000016@jgsu.edu.cn;Jie Li,E-mail:jie.li-2@manchester.ac.uk
Supported by:
The authors gratefully acknowledge the financial support from China Scholarship Council (CSC) (202406440073).

摘要/Abstract

摘要： Accurate prediction of pure component physiochemical properties is crucial for process integration, multiscale modelling, and optimization. In this work, an enhanced framework for pure component property prediction by using explainable machine learning methods is proposed. In this framework, the molecular representation method based on the connectivity matrix effectively considers atomic bonding relationships to automatically generate features. The supervised machine learning model random forest is applied for feature ranking and pooling. The adjusted R² is introduced to penalize the inclusion of additional features, providing an assessment of the true contribution of features. The prediction results for normal boiling point (T_b), liquid molar volume (L_mv), critical temperature (T_c) and critical pressure (P_c) obtained using Artificial Neural Network and Gaussian Process Regression models confirm the accuracy of the molecular representation method. Comparison with GC based models shows that the root-mean-square error on the test set can be reduced by up to 83.8%. To enhance the interpretability of the model, a feature analysis method based on Shapley values is employed to determine the contribution of each feature to the property predictions. The results indicate that using the feature pooling method reduces the number of features from 13316 to 100 without compromising model accuracy. The feature analysis results for T_b, L_mv, T_c, and P_c confirms that different molecular properties are influenced by different structural features, aligning with mechanistic interpretations. In conclusion, the proposed framework is demonstrated to be feasible and provides a solid foundation for mixture component reconstruction and process integration modelling.

关键词: Thermodynamic properties, Explainable machine learning, Molecular engineering, Shapley value, Adjusted R²

Abstract: Accurate prediction of pure component physiochemical properties is crucial for process integration, multiscale modelling, and optimization. In this work, an enhanced framework for pure component property prediction by using explainable machine learning methods is proposed. In this framework, the molecular representation method based on the connectivity matrix effectively considers atomic bonding relationships to automatically generate features. The supervised machine learning model random forest is applied for feature ranking and pooling. The adjusted R² is introduced to penalize the inclusion of additional features, providing an assessment of the true contribution of features. The prediction results for normal boiling point (T_b), liquid molar volume (L_mv), critical temperature (T_c) and critical pressure (P_c) obtained using Artificial Neural Network and Gaussian Process Regression models confirm the accuracy of the molecular representation method. Comparison with GC based models shows that the root-mean-square error on the test set can be reduced by up to 83.8%. To enhance the interpretability of the model, a feature analysis method based on Shapley values is employed to determine the contribution of each feature to the property predictions. The results indicate that using the feature pooling method reduces the number of features from 13316 to 100 without compromising model accuracy. The feature analysis results for T_b, L_mv, T_c, and P_c confirms that different molecular properties are influenced by different structural features, aligning with mechanistic interpretations. In conclusion, the proposed framework is demonstrated to be feasible and provides a solid foundation for mixture component reconstruction and process integration modelling.

Key words: Thermodynamic properties, Explainable machine learning, Molecular engineering, Shapley value, Adjusted R²

Jianfeng Jiao, Xi Gao, Jie Li. Pure component property estimation framework using explainable machine learning methods[J]. 中国化学工程学报, 2025, 84(8): 158-178.

Jianfeng Jiao, Xi Gao, Jie Li. Pure component property estimation framework using explainable machine learning methods[J]. Chinese Journal of Chemical Engineering, 2025, 84(8): 158-178.

参考文献

[1] P.J. Morris, Wetter is better for peat carbon, Nat. Clim. Chang. 11 (7) (2021) 561-562.
[2] Y. Yang, B. Chen, PSE in China: retrospect and prospects, Chem. Ind. Eng. Prog. 41 (2022) 3991-4008. (in Chinese).
[3] Wu Y, Molecular management for refining operations, Ph.D thesis,University of Manchester, 2010.
[4] D. Guan, L.Z. Zhang, Initial guess estimation and fast solving of petroleum complex molecular reconstruction model, AIChE. J. 68 (10) (2022) e17782.
[5] H.C. Gu, J.W. Li, P. Mu, Q.X. Zhu, Improving the operational efficiency of ethylene cracking integrated with refining by molecular management, Ind. Eng. Chem. Res. 59 (29) (2020) 13160-13174.
[6] E.T.C. Vogt, B.M. Weckhuysen, The refinery of the future, Nature 629 (8011) (2024) 295-306.
[7] V. Mann, R. Gani, V. Venkatasubramanian, Group contribution-based property modeling for chemical product design: a perspective in the AI era, Fluid Phase Equilib. 568 (2023) 113734.
[8] Y.X. Wei, L.Y. Shan, T. Qiu, D.N. Lu, Z. Liu, Machine learning-assisted retrosynthesis planning: current status and future prospects, Chin. J. Chem. Eng. 77 (2025) 273-292.
[9] J. Marrero, R. Gani, Group-contribution based estimation of pure component properties, Fluid Phase Equilib. 183 (2001) 183-208.
[10] A.S. Alshehri, A.K. Tula, F.Q. You, R. Gani, Next generation pure component property estimation models: With and without machine learning techniques, AIChE. J. 68 (6) (2022) e17469.
[11] R. Gani, Group contribution-based property estimation methods: advances and perspectives, Curr. Opin. Chem. Eng. 23 (2019) 184-196.
[12] H. Renon, Vapor-liquid equilibria using UNIFAC. A group contribution method, Fluid Phase Equilib. 1 (4) (1977) 317.
[13] D.S. Abrams, J.M. Prausnitz, Statistical thermodynamics of liquid mixtures: a new expression for the excess Gibbs energy of partly or completely miscible systems, AIChE. J. 21 (1) (1975) 116-128.
[14] A.R.N. Aouichaoui, F. Fan, J. Abildskov, G. Sin, Application of interpretable group-embedded graph neural networks for pure compound properties, Comput. Chem. Eng. 176 (2023) 108291.
[15] J. Burger, V. Papaioannou, S. Gopinath, G. Jackson, A. Galindo, C.S. Adjiman, A hierarchical method to integrated solvent and process design of physical CO₂ absorption using the SAFT-γ Mie approach, AIChE. J. 61 (10) (2015) 3249-3269.
[16] T. Zhou, Z. Song, X. Zhang, R. Gani, K. Sundmacher, Optimal solvent design for extractive distillation processes: a multiobjective optimization-based hierarchical framework, Ind. Eng. Chem. Res. 58 (15) (2019) 5777-5786.
[17] X.Y. Cao, M. Gong, A. Tula, X. Chen, R. Gani, V. Venkatasubramanian, An improved machine learning model for pure component property estimation, Engineering 39 (2024) 61-73.
[18] A.R.N. Aouichaoui, F. Fan, S.S. Mansouri, J. Abildskov, G. Sin, Combining Group-contribution concept and graph neural networks toward interpretable molecular property models, J. Chem. Inf. Model. 63 (3) (2023) 725-744.
[19] R. Gani, P.M. Harper, M. Hostrup, Automatic creation of missing groups through connectivity index for pure-component property prediction, Ind. Eng. Chem. Res. 44 (18) (2005) 7262-7269.
[20] D.P. Visco, R.S. Pophale, M.D. Rintoul, J.L. Faulon, Developing a methodology for an inverse quantitative structure-activity relationship using the signature molecular descriptor, J. Mol. Graph. Model. 20 (6) (2002) 429-438.
[21] D. Rogers, M. Hahn, Extended-connectivity fingerprints, J. Chem. Inf. Model. 50 (5) (2010) 742-754.
[22] M.R. Dobbelaere, Y. Ureel, F.H. Vermeire, L. Tomme, C.V. Stevens, K.M. Van Geem, Machine learning for physicochemical property prediction of complex hydrocarbon mixtures, Ind. Eng. Chem. Res. 61 (24) (2022) 8581-8594.
[23] W.M. Mi, H.J. Chen, D.A. Zhu, T. Zhang, F. Qian, Melting point prediction of organic molecules by deciphering the chemical structure into a natural language, Chem. Commun. 57 (21) (2021) 2633-2636.
[24] Z.N. Zeng, Y. Yao, Z.Y. Liu, M.S. Sun, A deep-learning system bridging molecule structure and biomedical text with comprehension comparable to human professionals, Nat. Commun. 13 (1) (2022) 862.
[25] G.B. Goh, N.O. Hodas, C. Siegel, A. Vishnu, SMILES2Vec: an interpretable general-purpose deep neural network for predicting chemical properties, 2019, https://doi.org/10.48550/arXiv.1712.02034.
[26] V. Mann, K. Brito, R. Gani, V. Venkatasubramanian, Hybrid, interpretable machine learning for thermodynamic property estimation using Grammar2vec for molecular representation, Fluid Phase Equilib. 561 (2022) 113531.
[27] O. Wieder, S. Kohlbacher, M. Kuenemann, A. Garon, P. Ducrot, T. Seidel, T. Langer, A compact review of molecular property prediction with graph neural networks, Drug Discov. Today Technol. 37 (2020) 1-12.
[28] B.Z. Dou, Z.L. Zhu, E. Merkurjev, L. Ke, L. Chen, J. Jiang, Y.Y. Zhu, J. Liu, B.G. Zhang, G.W. Wei, Machine learning methods for small data challenges in molecular science, Chem. Rev. 123 (13) (2023) 8736-8780.
[29] S. Ishida, T. Miyazaki, Y. Sugaya, S. Omachi, Graph neural networks with multiple feature extraction paths for chemical property estimation, Molecules 26 (11) (2021) 3125.
[30] X. Zang, X.B. Zhao, B.Z. Tang, Hierarchical molecular graph self-supervised learning for property prediction, Commun. Chem. 6 (1) (2023) 34.
[31] S. Jain, B.C. Wallace, Attention is not explanation, ArXiv Prepr. ArXiv1902.10186 (2019).
[32] S. Wiegreffe, Y. Pinter, Attention is not not explanation, ArXiv Prepr. ArXiv1908.04626 (2019).
[33] B. Rozemberczki, L. Watson, P. Bayer, H.T. Yang, O. Kiss, S. Nilsson, R. Sarkar, The shapley value in machine learning,In:Proceedings of the Thirty-First International Joint Conference on Artificial Intelligence. Vienna, Austria., 2022.
[34] A. Yang, S.R. Sun, Y. Su, Z.Y. Kong, J.Z. Ren, W.F. Shen, Insight to the prediction of CO₂ solubility in ionic liquids based on the interpretable machine learning model, Chem. Eng. Sci. 297 (2024) 120266.
[35] Q. Pan, X.L. Fan, J. Li, Automatic creation of molecular substructures for accurate estimation of pure component properties using connectivity matrices, Chem. Eng. Sci. 265 (2023) 118214.
[36] K.X. Bi, T. Qiu, An intelligent SVM modeling process for crude oil properties prediction based on a hybrid GA-PSO method, Chin. J. Chem. Eng. 27 (8) (2019) 1888-1894.
[37] G.S.K. Ranjan, A. Kumar Verma, S. Radhika, K-Nearest neighbors and grid search CV based real time fault monitoring system for industries, In:2019 IEEE 5th International Conference for Convergence in Technology (I2CT). Bombay, India. IEEE, 2019.
[38] Y. Wang, P.X. Wang, K. Tansey, J.M. Liu, B. Delaney, W.T. Quan, An interpretable approach combining Shapley additive explanations and LightGBM based on data augmentation for improving wheat yield estimates, Comput. Electron. Agric. 229 (2025) 109758.
[39] Y. Shi, W.M. Zhong, X. Peng, M.L. Yang, W. Du, Interpretable reconstruction of naphtha components using property-based extreme gradient boosting and compositional-weighted Shapley additive explanation values, Chem. Eng. Sci. 284 (2024) 119462.
[40] G. Landrum, Rdkit documentation, Release 1 (2013) 4.
[41] M. Greenacre, P.J.F. Groenen, T. Hastie, A.I. D’Enza, A. Markos, E. Tuzhilina, Principal component analysis, Nat. Rev. Meth. Primers 2 (2022) 100.
[42] Y.S. Mu, X.D. Liu, L.D. Wang, A Pearson’s correlation coefficient based decision tree and its parallel implementation, Inf. Sci. 435 (2018) 40-58.
[43] S.J. Rigatti, Random forest, J. Insur. Med. 47 (1)31–39.
[44] M. Schonlau, R.Y. Zou, The random forest algorithm for statistical learning, Stata J. Promot. Commun. Stat. Stata 20 (1) (2020) 3-29.
[45] F. Jin, Y. Guo, Y. Zhang, X. Ma, B. Liu, J. Jiao, X. Yin, H. Xu, J. Gong, K. Wen, A new gas pipeline network simulation method based on BHC-PINN, Nat. Gas Ind. 45 (2025) 164-174.
[46] Z.Q. Wang, D.K. He, H.T. Nie, Operational optimization of copper flotation process based on the weighted Gaussian process regression and index-oriented adaptive differential evolution algorithm, Chin. J. Chem. Eng. 66 (2024) 167-179.
[47] A. Gulli, S. Pal, Deep Learning with Keras, Packt Publishing Ltd., 2017.
[48] B. Pang, E. Nijkamp, Y.N. Wu, Deep learning with TensorFlow: a review, J. Educ. Behav. Stat. 45 (2) (2020) 227-248.
[49] X. Yin, K. Wen, Y. Wu, X. Han, Y. Mukhtar, J. Gong, A machine learning-based surrogate model for the rapid control of piping flow: Application to a natural gas flowmeter calibration system, J. Nat. Gas Sci. Eng. 98 (2022) 104384.
[50] X. Yin, K. Wen, W.H. Huang, Y.W. Luo, Y. Ding, J. Gong, J.F. Gao, B.Y. Hong, A high-accuracy online transient simulation framework of natural gas pipeline network by integrating physics-based and data-driven methods, Appl. Energy 333 (2023) 120615.
[51] K. Wen, J.F. Jiao, K. Zhao, X. Yin, Y. Liu, J. Gong, C.C. Li, B.Y. Hong, Rapid transient operation control method of natural gas pipeline networks based on user demand prediction, Energy 264 (2023) 126093.
[52] O. Kramer, Scikit-learn. Machine Learning for Evolution Strategies. Springer International Publishing, (2016), pp 5-53.
[53] M. Sundararajan, A. Najmi, The many shapley values for model explanation, International In:Conference on Machine Learning, online, 2019.
[54] H.G. Cai, Y. Yang, Y.Q. Tang, Z.Y. Sun, W.S. Zhang, Shapley value-based class activation mapping for improved explainability in neural networks, Vis. Comput. (2025) https://doi.org/10.1007/s00371-025-03803-1.
[55] G. Modla, Energy saving methods for the separation of a minimum boiling point azeotrope using an intermediate entrainer, Energy 50 (2013) 103-109.
[56] D.B. Robinson, D.Y. Peng, S.Y. Chung, The development of the Peng - Robinson equation and its application to phase equilibrium in a system containing methanol, Fluid Phase Equilib. 24 (1-2) (1985) 25-41.
[57] L. Constantinou, R. Gani, J.P. O’Connell, Estimation of the acentric factor and the liquid molar volume at 298 K using a new group contribution method, Fluid Phase Equilib. 103 (1) (1995) 11-22.
[58] Y.C. Chen, A tutorial on kernel density estimation and recent advances, biostat epidemiol 1 (1) (2017) 161-187.
[59] C.E. Shannon, A mathematical theory of communication, Bell Syst. Tech. J. 27 (3) (1948) 379-423.
[60] L. Prechelt, Automatic early stopping using cross validation: quantifying the criteria, Neural Netw. 11 (4) (1998) 761-767.
[61] A.S. Alshehri, A.K. Tula, F. You, R. Gani, Corrections to “Next generation pure component property estimation models: With and without machine learning techniques,” AIChE J. 69 (2023) e18086.
[62] Z.H. Zhou, Machine Learning. Springer Singapore, 2021.
[63] W.C. Edmister, Thermodynamic properties of hydrocarbons, Ind. Eng. Chem. 30 (3) (1938) 352-358.

Pure component property estimation framework using explainable machine learning methods

Pure component property estimation framework using explainable machine learning methods

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 15

编辑推荐

Metrics

本文评价

[1]	Wenting Cheng, Qianqian Li, Ying Zhai, Huaigang Cheng, Fangqin Cheng. Micellization behavior and thermodynamic properties of cetyltrimethylammonium bromide in lithium chloride, potassium chloride, magnesium chloride and calcium chloride solutions[J]. 中国化学工程学报, 2025, 81(5): 95-104.
[2]	Shan Feng, Maolan Li, Xudong Yu, Lin Wang, Qin Huang. Phase equilibria of ternary system RbCl - polyethylene glycol (PEG6000) - H₂O at T = (288.2, 298.2, and 308.2) K: Measurement, correlation, and thermodynamic modeling[J]. 中国化学工程学报, 2025, 81(5): 105-114.
[3]	Hadi Taheri Parsa, Hossein Iloukhani, Khatereh Khanlarzadeh. Experimental and computational insight in thermodynamic properties of binary mixtures of acetonitrile with trichloroethene or tetrachloroethene at different temperatures[J]. 中国化学工程学报, 2025, 80(4): 328-340.
[4]	Wensheng Li, Liangyuan Qi, Daolin Ye, Wei Cai, Weiyi Xing. Facile modification of aluminum hypophosphate and its flame retardancy for polystyrene[J]. 中国化学工程学报, 2023, 60(8): 90-98.
[5]	Huiqi Wang, Jianpo Ren, Shihao Zhang, Jiayu Dai, Yue Niu, Ketao Shi, Qiuxiang Yin, Ling Zhou. Measurement and correlation of solubility of 9-fluorenone in 11 pure organic solvents from T = 283.15 to 323.15 K[J]. 中国化学工程学报, 2023, 60(8): 235-241.
[6]	Li Xia, Yule Pan, Tingting Zhao, Xiaoyan Sun, Shaohui Tao, Yushi Chen, Shuguang Xiang. Estimating heat capacities of liquid organic compounds based on elements and chemical bonds contribution[J]. 中国化学工程学报, 2023, 57(5): 30-38.
[7]	Arnop Dutta, Md. Tuhinur R. Joy, Sk. Md. Ali Ahsan, Mansour K. Gatasheh, Dileep Kumar, Malik Abdul Rub, Md. Anamul Hoque, Mohammad Majibur Rahman, Nasrul Hoda, D.M. Shafiqul Islam. Physico-chemical parameters for the assembly of moxifloxacin hydrochloride and cetyltrimethylammonium chloride mixture in aqueous and alcoholic media[J]. 中国化学工程学报, 2023, 57(5): 280-289.
[8]	Peng Yang, Shengzhe Jia, Yan Wang, Zongqiu Li, Songgu Wu, Jingkang Wang, Junbo Gong. Dissolution behavior, thermodynamic and kinetic analysis of malonamide by experimental measurement and molecular simulation[J]. 中国化学工程学报, 2023, 53(1): 260-269.
[9]	Yuanjie Li, Qiuxiang Yin, Meijing Zhang, Ying Bao, Baohong Hou, Jingkang Wang, Jiting Huang, Ling Zhou. Characterization and structure analysis of the heterosolvate of erythromycin thiocyanate[J]. 中国化学工程学报, 2022, 44(4): 268-274.
[10]	Nuochen Zhang, Yuande Dai, Linghao Feng, Biao Li. Study on environmentally friendly refrigerant R13I1/R152a as an alternative for R134a in automotive air conditioning system[J]. 中国化学工程学报, 2022, 44(4): 292-299.
[11]	Zhengyu Chen, Xinhui Yao, Dong Guan, Suoqi Zhao, Linzhou Zhang, Chunming Xu. Vacuum residue coking process simulation using molecular-level kinetic model coupled with vapor-liquid phase separation[J]. 中国化学工程学报, 2022, 41(1): 301-310.
[12]	Huan Zhou, Peng Wu, Wenxuan Li, Xingfan Wang, Kuo Zhou, Qing Hao. Thermodynamic modeling and phase diagram prediction of salt lake brine systems II. Aqueous Li⁺-Na⁺-K⁺-SO₄^2- and its subsystems[J]. 中国化学工程学报, 2021, 34(6): 134-149.
[13]	Jiaqi Ding, Nan Xu, Manh Tien Nguyen, Qi Qiao, Yao Shi, Yi He, Qing Shao. Machine learning for molecular thermodynamics[J]. 中国化学工程学报, 2021, 29(3): 227-239.
[14]	Rui Wu, Ya-Ping Wang, Lin Shao, Wei Wang, Bi-Yu Tang. Thermodynamic property of ternary compound MgCaSi: A study from ab initio Debye-Grüneisen model[J]. 中国化学工程学报, 2021, 40(12): 315-322.
[15]	Yanmin Shen, Wenju Liu, Paifeng Shi, Chao Wang. Solubility measurement and thermodynamic properties of sulfamonomethoxine in pure solvents and sulfamonomethoxine hydrate in acetone + water binary solvent at different temperature[J]. 中国化学工程学报, 2021, 38(10): 196-204.