A machine learning model for pK<sub>a</sub> prediction and generalization of organic compounds in ionic liquids

WANG Chenyang; XIAO Jianwei; PENG Kuilin; YANG Zhen

doi:10.11949/0438-1157.20260112

您当前的位置：

首页 >

文章列表页 >

A machine learning model for pK_a prediction and generalization of organic compounds in ionic liquids

更新时间：2026-06-26

- A machine learning model for pK_a prediction and generalization of organic compounds in ionic liquids
- CIESC Journal (2026)
- 作者机构：
  
  江西师范大学化学工程学院，江西南昌 330022
- 作者简介：
- 基金信息：
- DOI：10.11949/0438-1157.20260112
  CLC： O 645.1
- Received：23 January 2026，
  
  Revised：2026-06-25，
  
  Accepted：26 June 2026，
- 稿件说明：
移动端阅览
WANG Chenyang, XIAO Jianwei, PENG Kuilin, et al. A machine learning model for pK_a prediction and generalization of organic compounds in ionic liquids[J/OL]. CIESC Journal, 2026. DOI: 10.11949/0438-1157.20260112.
DOI：

WANG Chenyang, XIAO Jianwei, PENG Kuilin, et al. A machine learning model for pK_a prediction and generalization of organic compounds in ionic liquids[J/OL]. CIESC Journal, 2026. DOI: 10.11949/0438-1157.20260112. DOI：

摘要

有机化合物在离子液体中的解离常数（pK

）是关键物化参数，其实验测定成本高、周期长，而现有理论模型普遍存在跨溶剂泛化能力不足的问题。为此，本研究构建了两种机器学习预测框架，系统评估其性能边界与应用场景。首先，结合图神经网络（GNN）与机器学习算法（ML），利用GNN自动提取分子图特征，再输入至ML模型预测。在已知溶剂体系中，消息传递神经网络（MPNN）与梯度提升回归树（GBRT）组合模型表现最优，测试集R²达0.9822。进一步采用“留一溶剂”策略考察跨溶剂泛化能力，发现GNN+ML框架在该场景下效果不稳定。为此，本研究引入基于RDKit的分子描述符，结合机器学习构建跨溶剂自适应迁移学习模型。结果表明，基于RDKit描述符的模型在样本充足条件下能有效实现跨溶剂预测，其中原子对指纹（AP_FP）和RDKit指纹（RD_FP）与XGBoost或GBRT组合时，在多数测试集上R²超过0.95，展现出优异泛化性能。本研究为五种离子液体中pK

的高通量预测与溶剂设计提供了有效的计算工具。

Abstract

The dissociation constant (pK

) of organic compounds in ionic liquids (ILs) is a critical physicochemical property for understanding their behavior in these versatile and green solvent systems

yet experimental measurements remain costly and labor-intensive

and existing theoretical models often lack generalization across diverse solvent environments

particularly within the vast chemical space of ILs. To overcome these challenges

this study developed and systematically compared two distinct machine learning (ML) frameworks to evaluate their performance boundaries and applicable scenarios. First

a hybrid framewo

rk integrating Graph Neural Networks (GNNs)—specifically Message Passing Neural Networks (MPNN) and Graph Attention Networks (GAT)—with traditional ML algorithms was constructed to automatically extract topological and bond-level features from molecular graphs of both organic compounds and IL ions

which were then fed into six ML models including XGBoost

GBRT

Bagging

and ET; under an 80/20 train-test split within known IL systems

the MPNN+GBRT model achieved the best prediction performance with a test-set R

of 0.9822

MAE of 0.351

and RMSE of 0.524

demonstrating high in-distribution accuracy. To critically assess generalization to unseen ILs

a cross-solvent prediction task using a leave-one-solvent-out strategy revealed that the GNN+ML framework

represented by the optimal MPNN+GBRT model

exhibited unstable performance with the highest R

reaching only 0.9331 and showing strong dependence on training data ion structures. Given this limitation

an alternative approach employing RDKit-based molecular descriptors—including Atom Pair

Morgan

RDKit

and Torsion fingerprints—as feature inputs for ML models was pursued

and when evaluated on the same cross-solvent task

these descriptor-based models successfully achieved effective cross-solvent prediction under sufficient sample availability. In summary

this study underscores that the choice of feature representation is crucial for balancing predictive accuracy and generalization in IL pK

modeling

with the GNN+ML framework excelling in high-accuracy predictions within known chemical environments while the RDKit-descriptor-based ML approach offers a robust and generalizable solution for novel IL systems; this work not only provides two computational tools for high-throughput pK

prediction and solvent screening across five ILs but also delivers methodological insights into feature engineering for property prediction in tunable solvent systems

with future efforts directed toward expanding the IL dataset

ploring hybrid feature strategies

and improving model interpretability.

关键词

Keywords

references

Hammett L P . Some relations between reaction rates and equilibrium constants [J ] . Chemical Reviews , 1935 , 17 ( 1 ): 125 - 136 .

Yang J D , Ji P J , Xue X S , et al . Recent advances and advisable applications of bond energetics in organic chemistry [J ] . Journal of the American Chemical Society , 2018 , 140 ( 28 ): 8611 - 8623 .

于海瀛 , 陈伟 , 袁泉 , 等 . 有机小分子化合物解离常数pKa的预测研究进展 [J ] . 科学通报 , 2015 , 60 ( 14 ): 1261-1271, 1 - 3 .

Yu H Y , Chen W , Yuan Q , et al . A review on predicting pKa values of small organic compounds [J ] . Chinese Science Bulletin , 2015 , 60 ( 14 ): 1261-1271, 1 - 3 .

Bors D A , Kaufman M J , Streitwieser Jr A . Carbon acidity. 67. the indicator scale of cesium ion pairs in tetrahydrofuran [J ] . Journal of the American Chemical Society , 1985 , 107 ( 24 ): 6975 - 6982 .

Lambert C , von Ragué Schleyer P . Are polar organometallic compounds “carbanions” The gegenion effect on structure and energies of alkali-metal compounds [J ] . Angewandte Chemie International Edition , 1994 , 33 ( 11 ): 1129 - 1140 .

Rupp M , Korner R , Tetko I V . Predicting the pKa of small molecules [J ] . Combinatorial Chemistry & High Throughput Screening , 2011 , 14 ( 5 ): 307 - 327 .

Matthews W S , Bares J E , Bartmess J E , et al . Equilibrium acidities of carbon acids. VI. Establishment of an absolute scale of acidities in dimethyl sulfoxide solution [J ] . Journal of the American Chemical Society , 1975 , 97 ( 24 ): 7006 - 7014 .

Coetzee J F , Kolthoff I M . Polarography in acetonitrile. III. Brønsted acids. amperometric titration of amines with perchloric acid. oxygen [J ] . Journal of the American Chemical Society , 1957 , 79 ( 23 ): 6110 - 6115 .

Welton T . Ionic liquids: a brief history [J ] . Biophysical Reviews , 2018 , 10 ( 3 ): 691 - 706 .

Shamshina J L , Rogers R D . Ionic liquids: new forms of active pharmaceutical ingredients with unique, tunable properties [J ] . Chemical Reviews , 2023 , 123 ( 20 ): 11894 - 11953 .

Nasirpour N , Mohammadpourfard M , Zeinali Heris S . Ionic liquids: Promising compounds for sustainable chemical processes and applications [J ] . Chemical Engineering Research and Design , 2020 , 160 : 264 - 300 .

Reijenga J , van Hoof A , van Loon A , et al . Development of methods for the determination of pKa Values [J ] . Analytical Chemistry Insights , 2013 , 8: ACI.S 12304 .

Webb H , Tynan-Connolly B M , Lee G M , et al . Remeasuring HEWL pKa values by NMR spectroscopy: Methods, analysis, accuracy, and implications for theoretical pKa calculations [J ] . Proteins: Structure, Function, and Bioinformatics , 2011 , 79 ( 3 ): 685 - 702 .

Hu Y H , Xing Y Y , Yue H , et al . Ionic liquids revolutionizing biomedicine: recent advances and emerging opportunities [J ] . Chemical Society Reviews , 2023 , 52 ( 20 ): 7262 - 7293 .

Andonegi M , Correia D , Pereira N , et al . Sustainable collagen blends with different ionic liquids for resistive touch sensing applications [J ] . ACS Sustainable Chemistry & Engineering , 2023 , 11 ( 15 ): 5986 - 5998 .

Clark J , Perrin D D . Prediction of the strengths of organic bases [J ] . Quarterly Reviews, Chemical Society , 1964 , 18 ( 3 ): 295 - 320 .

Perrin D D , Dempsey B , Serjeant E P . PKa Prediction for Organic Acids and Bases [M ] . London : Chapman and Hall , 1981 .

Lee A C . SMARTS Approach to chemical data mining and physicochemical property prediction [M ] . University of Michigan , 2009 .

Lee P H , Ayyampalayam S N , Carreira L A , et al . In silico prediction of ionization constants of drugs [J ] . Molecular Pharmaceutics , 2007 , 4 ( 4 ): 498 - 512 .

Agrawal A , Deshpande P D , Cecen A , et al . Exploration of data science techniques to predict fatigue strength of steel from composition and processing parameters [J ] . Integrating Materials and Manufacturing Innovation , 2014 , 3 ( 1 ): 90 - 108 .

Yamada H , Liu C , Wu S , et al . Predicting materials properties with little data using shotgun transfer learning [J ] . ACS Central Science , 2019 , 5 ( 10 ): 1717 - 1730 .

Zhu M Y , Lin Y T , Pang Z Y , et al . An automated workflow for lithium metal oxide screening assisted by large language models [J ] . Chinese Chemical Letters , 2025 : 112217 .

Gao X , Yang A Q , Yu W B , et al . Generative artificial intelligence navigated development of solvents for next generation high-performance magnesium batteries [J ] . Advanced Materials , 2026 , 38 ( 1 ): e10083 .

Wu Z H , Pan S R , Chen F W , et al . A comprehensive survey on graph neural networks [J ] . IEEE Transactions on Neural Networks and Learning Systems , 2021 , 32 ( 1 ): 4 - 24 .

Roszak R , Beker W , Molga K , et al . Rapid and accurate prediction of pKa values of C-H acids using graph convolutional neural networks [J ] . Journal of the American Chemical Society , 2019 , 141 ( 43 ): 17142 - 17149 .

Pan X L , Wang H , Li C Y , et al . MolGpk a : a web server for small molecule pKa prediction using a graph-convolutional neural network [J ] . Journal of Chemical Information and Modeling , 2021 , 61 ( 7 ): 3159 - 3165 .

Xiong J C , Li Z J , Wang G C , et al . Multi-instance learning of graph neural networks for aqueous pKa prediction [J ] . Bioinformatics , 2022 , 38 ( 3 ): 792 - 798 .

Chen T , He T , Benesty M , et al . Xgboost: extreme gradient boosting [J ] . R package version 0.4-2 , 2015 , 1 ( 4 ): 1 - 4 .

Cutler A , Cutler D R , Stevens J R . Random forests [M ] . Ensemble machine learning . Springer , New York, NY , 2012 : 157 - 175 .

Mansouri K , Cariello N F , Korotcov A , et al . Open-source QSAR models for pKa prediction using multiple machine learning approaches [J ] . Journal of Cheminformatics , 2019 , 11 ( 1 ): 60 .

Wu J L , Kang Y , Pan P C , et al . Machine learning methods for pKa prediction of small molecules: Advances and challenges [J ] . Drug Discovery Today , 2022 , 27 ( 12 ): 103372 .

Yang Q , Li Y , Yang J D , et al . Holistic prediction of the pKa in diverse solvents based on a machine-learning approach [J ] . Angewandte Chemie International Edition , 2020 , 59 ( 43 ): 19282 - 19291 .

Brown S J , Yalcin D , Pandiancherri S , et al . Characterising a protic ionic liquid library with applied machine learning algorithms [J ] . Journal of Molecular Liquids , 2022 , 367 : 120453 .

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Machine learning drives system optimization of liquid organic hydrogen storage technology

Predicting and interpreting the toxicity of ionic liquids using graph neural network

Prediction of biomass pyrolysis product yield based on machine learning algorithms

Advantages and risks of swirling cooling granulation towers

Oxygen mass transfer performance of hollow fiber non-bubble aeration membrane based on gas/liquid-phase measurement method

Related Author

FENG Haijun

ZHANG Bingxuan

ZHOU Jian

FAN Xiayu

SUN Jianchen

LI Keying

YAO Xinya

SHANG Hui

Related Institution

School of Computer Sciences, Shenzhen Institute of Information Technology

School of Chemistry and Chemical Engineering, Guangdong Provincial Key Laboratory for Green Chemical Product Technology, South China University of Technology

College of Chemical Engineering and Environment, China University of Petroleum

College of Artificial Intelligence, China University of Petroleum

Institute of Agricultural Environment and Sustainable Development, Chinese Academy of Agricultural Sciences

⁰

A machine learning model for pKa prediction and generalization of organic compounds in ionic liquids

DOI：10.11949/0438-1157.20260112

摘要

Abstract

关键词

Keywords

references

A machine learning model for pK_a prediction and generalization of organic compounds in ionic liquids