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

doi:10.11949/0438-1157.20240663

Abstract

Abstract:

Ionic liquids are potentially toxic to the environment, how to control the toxicity of ionic liquids is one of the key factors. To understand their toxicity mechanisms, three traditional machine learning methods (support vector machine, random forest, multilayer perceptron) and three graph neural network models (graph attention network, message passing neural network, graph convolutional model) were established to predict the toxicity of ionic liquids in four living organisms (leukemia rat cell line IPC-81, acetylcholinesterase, Escherichia coli, and Vibrio fischeri). The simplified molecular-input line-entry system (SMILES) of molecules and toxicity lgEC₅₀ values work as the input and output respectively. In the three traditional machine learning methods, extended-connectivity fingerprints (ECFPs) were used to represent molecules. While in the three graph neural network models, molecular graphs were used to represent molecules. Benefiting from molecular structure information, the graph convolutional model (GCM) had lower RMSE and MAE, and higher R² than other models in all four datasets. Therefore, the GCM model was superior in predicting the toxicity of ionic liquids. Meanwhile, based on the GCM model, an intepretability model was established to analyze the contribution of atomic groups to the toxicity of ionic liquids in a data-driven procedure. The aromatic ring of cations and long alkyl chain could produce toxicity. Atomic groups such as S⁺, P⁺, N⁺, and NH⁺ could significantly enhance the toxicity of ionic liquids, while atomic groups such as P^-, F, B^-, and C could effectively reduce the toxicity of ionic liquids. This discovery provides a theoretical basis for rapid screening and development of greener and low-toxicity ionic liquids.

Key words: ionic liquids, toxicity, machine learning, graph neural network, model, prediction, interpretability

摘要：

离子液体对环境有潜在毒性，为了解其毒性机制，建立了三种传统机器学习（支持向量机，随机森林，多层感知机）和三种图神经网络（图注意力网络，消息传递神经网络，图卷积模型）模型，预测离子液体对大鼠IPC-81细胞等4种活生物体的毒性。凭借分子结构信息，图卷积模型在4个数据集中的RMSE和MAE均最低，R²均最高，因此，图卷积模型在预测离子液体毒性上更优越。同时，基于图卷积模型，建立毒性解释模型，从数据驱动上来分析原子基团对毒性的贡献。阳离子的芳香环和长烷基链会产生毒性，S⁺、P⁺、N⁺、NH⁺等原子基团会显著增强离子液体的毒性，而P^-、F、B^-、C等原子基团会有效降低离子液体的毒性。该发现可为快速筛选和开发更绿色低毒型离子液体提供理论依据。

关键词: 离子液体, 毒性, 机器学习, 图神经网络, 模型, 预测, 可解释性

CLC Number:

O 645.1

Haijun FENG, Bingxuan ZHANG, Jian ZHOU. Predicting and interpreting the toxicity of ionic liquids using graph neural network[J]. CIESC Journal, 2025, 76(1): 93-106.

冯海军, 章冰璇, 周健. 图神经网络模型预测和解释离子液体毒性的研究[J]. 化工学报, 2025, 76(1): 93-106.

Figures/Tables 12

References 42

1	Zhu M Y, Yang Y. Poly(ionic liquid)s: an emerging platform for green chemistry[J]. Green Chemistry, 2024, 26(9): 5022-5102.
2	张璐, 刘杰, 娄生辉, 等. 季膦萘磺酸盐离子液体与甲基膦酸二甲酯协同阻燃环氧树脂[J]. 高等学校化学学报, 2024, 45(4): 17-26.
	Zhang L, Liu J, Lou S H, et al. Synergistic flame-retardant epoxy resin using quaternary phosphate naphthalene sulfonate ionic liquid and dimethyl methylphosphonate[J]. Chemical Journal of Chinese Universities, 2024, 45(4): 17-26.
3	Probert P M, Leitch A C, Dunn M P, et al. Identification of a xenobiotic as a potential environmental trigger in primary biliary cholangitis[J]. Journal of Hepatology, 2018, 69(5): 1123-1135.
4	Zhang C, Zhu L S, Wang J H, et al. The acute toxic effects of imidazolium-based ionic liquids with different alkyl-chain lengths and anions on zebrafish (Danio rerio)[J]. Ecotoxicology and Environmental Safety, 2017, 140: 235-240.
5	Jeremias G, Jesus F, Ventura S P M, et al. New insights on the effects of ionic liquid structural changes at the gene expression level: molecular mechanisms of toxicity in daphnia Magna[J]. Journal of Hazardous Materials, 2021, 409: 124517.
6	Jakubowska A, Grabińska-Sota E. Study of the toxicity of five quaternary ammonium ionic liquids to aquatic organisms[J]. Desalination and Water Treatment, 2018, 117: 202-210.
7	Makarov D M, Fadeeva Y A, Safonova E A, et al. Predictive modeling of antibacterial activity of ionic liquids by machine learning methods[J]. Computational Biology and Chemistry, 2022, 101: 107775.
8	Feng H J, Qin L L, Zhang B X, et al. Prediction and interpretability of melting points of ionic liquids using graph neural networks[J]. ACS Omega, 2024, 9(14): 16016-16025.
9	Venkatraman V, Evjen S, Knuutila H K, et al. Predicting ionic liquid melting points using machine learning[J]. Journal of Molecular Liquids, 2018, 264: 318-326.
10	Feng H J, Qin W, Hu G W, et al. Intelligent prediction of nitrous oxide capture in designable ionic liquids[J]. Applied Sciences, 2023, 13(12): 6900.
11	Feng H J, Zhang P G, Qin W, et al. Estimation of solubility of acid gases in ionic liquids using different machine learning methods[J]. Journal of Molecular Liquids, 2022, 349: 118413.
12	Jian Y, Wang Y Y, Barati Farimani A. Predicting CO₂ absorption in ionic liquids with molecular descriptors and explainable graph neural networks[J]. ACS Sustainable Chemistry & Engineering, 2022, 10(50): 16681-16691.
13	Makarov D M, Fadeeva Y A, Golubev V A, et al. Designing deep eutectic solvents for efficient CO₂ capture: a data-driven screening approach[J]. Separation and Purification Technology, 2023, 325: 124614.
14	Song C C, Wang C Y, Fang F, et al. Large-scale screening for high conductivity ionic liquids via machine learning algorithm utilizing graph neural network-based features[J]. Journal of Chemical & Engineering Data, DOI: 10.1021/acs.jced.3c00709 .
15	肖拥君, 时兆翀, 万仁, 等. 反向传播神经网络用于预测离子液体的自扩散系数[J]. 化工学报, 2024, 75(2): 429-438.
	Xiao Y J, Shi Z C, Wan R, et al. Prediction of self-diffusion coefficients of ionic liquids using back-propagation neural networks[J]. CIESC Journal, 2024, 75(2): 429-438.
16	曾少娟, 孙雪琦, 白银鸽, 等. CO₂捕集分离的功能离子液体及材料研究进展[J]. 化学学报, 2023, 81(6): 627-645.
	Zeng S J, Sun X Q, Bai Y G, et al. Research progress of CO₂ capture and separation by functionalized ionic liquids and materials[J]. Acta Chimica Sinica, 2023, 81(6): 627-645.
17	Abdullah N H, Zaini D, Lal B. Prediction of ionic liquids toxicity using machine learning models for application to gas hydrate[J]. Process Safety Progress, 2024, 43(S1): S199-S212.
18	赵永升, 黄莺, 赵继红, 等. 基于QSAR方法的离子液体毒性预测[J]. 化工学报, 2014, 65(5): 1616-1621.
	Zhao Y S, Huang Y, Zhao J H, et al. QSAR for predicting toxicity of imidazolium-based ionic liquids[J]. CIESC Journal, 2014, 65(5): 1616-1621.
19	Ma S Y, Lv M, Zhang X Y, et al. Computational study of the effects of cations and anions to the cytotoxicity of diverse ionic liquids by supervised machine learning[J]. Chemometrics and Intelligent Laboratory Systems, 2015, 144: 138-147.
20	Sosnowska A, Grzonkowska M, Puzyn T. Global versus local QSAR models for predicting ionic liquids toxicity against IPC-81 leukemia rat cell line: the predictive ability[J]. Journal of Molecular Liquids, 2017, 231: 333-340.
21	Zhao Y S, Zhao J H, Huang Y, et al. Toxicity of ionic liquids: database and prediction via quantitative structure-activity relationship method[J]. Journal of Hazardous Materials, 2014, 278: 320-329.
22	Cao L D, Zhu P, Zhao Y S, et al. Using machine learning and quantum chemistry descriptors to predict the toxicity of ionic liquids[J]. Journal of Hazardous Materials, 2018, 352: 17-26.
23	Yan J C, Liu G H, Chen H L, et al. ILTox: a curated toxicity database for machine learning and design of environmentally friendly ionic liquids[J]. Environmental Science & Technology Letters, 2023, 10(11): 983-988.
24	Yan J C, Yan X L, Hu S, et al. Comprehensive interrogation on acetylcholinesterase inhibition by ionic liquids using machine learning and molecular modeling[J]. Environmental Science & Technology, 2021, 55(21): 14720-14731.
25	Ahmad W, Tayara H, Chong K T. Attention-based graph neural network for molecular solubility prediction[J]. ACS Omega, 2023, 8(3): 3236-3244.
26	Rittig J G, Ben Hicham K, Schweidtmann A M, et al. Graph neural networks for temperature-dependent activity coefficient prediction of solutes in ionic liquids[J]. Computers & Chemical Engineering, 2023, 171: 108153.
27	Atz K, Grisoni F, Schneider G. Geometric deep learning on molecular representations[J]. Nature Machine Intelligence, 2021, 3: 1023-1032.
28	Wang Y Y, Magar R, Liang C, et al. Improving molecular contrastive learning via faulty negative mitigation and decomposed fragment contrast[J]. Journal of Chemical Information and Modeling, 2022, 62(11): 2713-2725.
29	Wang Y Y, Wang J R, Cao Z L, et al. Molecular contrastive learning of representations via graph neural networks[J]. Nature Machine Intelligence, 2022, 4: 279-287.
30	Karamad M, Magar R, Shi Y T, et al. Orbital graph convolutional neural network for material property prediction[J]. Physical Review Materials, 2020, 4(9): 093801.
31	Duvenaud D, Maclaurin D, Aguilera-Iparraguirre J, et al. Convolutional networks on graphs for learning molecular fingerprints[J]. Advances in Neural Information Processing Systems, 2015, 28: 2224-2232.
32	Kearnes S, McCloskey K, Berndl M, et al. Molecular graph convolutions: moving beyond fingerprints[J]. Journal of Computer-Aided Molecular Design, 2016, 30(8): 595-608.
33	Veličković P, Cucurull G, Casanova A, et al. Graph attention networks[EB/OL]. 2017. .
34	Gilmer J, Schoenholz S S, Riley P F, et al. Neural message passing for quantum chemistry[EB/OL]. 2017. .
35	Fan R, Chang K, Hsieh C, et al. Liblinear: a library for large linear classification[J]. The Journal of Machine Learning Research, 2008, 9: 1871-1874.
36	Breiman L. Random forests[J]. Machine Learning, 2001, 45: 5-32.
37	Geurts P, Ernst D, Wehenkel L. Extremely randomized trees[J]. Machine Learning, 2006, 63(1): 3-42.
38	Hinton G E. Connectionist learning procedures[M]//Mechine Learning: An Artificial Intelligence Approach Volume Ⅲ. Elsevier, 1990: 555-610.
39	Ramsundar B, Eastman P, Walters P, et al. Deep Learning for the Life Sciences: Applying Deep Learning to Genomics, Microscopy, Drug Discovery, and More[M]. Sebastopol, CA: O'Reilly Media, 2019.
40	Pedregosa F, Varoquaux G, Gramfort A, et al. Scikit-learn: machine learning in python[J]. The Journal of Machine Learning Research, 2011, 12: 2825-2830.
41	Buitinck L, Louppe G, Blondel M, et al. API design for machine learning software: experiences from the scikit-learn project[EB/OL]. 2013. .
42	Weyhing-Zerrer N, Gundolf T, Kalb R, et al. Predictability of ionic liquid toxicity from a SAR study on different systematic levels of pathogenic bacteria[J]. Ecotoxicology and Environmental Safety, 2017, 139: 394-403.

数据集	评估指标	SVM	RF	MLP	GAT	MPNN	GCM
大鼠白血病细胞（IPC-81）	RMSE	0.687	0.655	0.640	0.519	0.486	0.457
	MAE	0.510	0.436	0.467	0.379	0.353	0.325
	R²	0.683	0.706	0.688	0.764	0.787	0.867
对乙酰胆碱酯酶（AChE）	RMSE	0.229	0.286	0.243	0.295	0.246	0.229
	MAE	0.128	0.170	0.156	0.210	0.171	0.184
	R²	0.814	0.746	0.825	0.716	0.822	0.830
大肠杆菌（E. coli）	RMSE	0.545	0.547	0.665	0.648	0.610	0.396
	MAE	0.396	0.407	0.462	0.471	0.417	0.324
	R²	0.832	0.811	0.775	0.760	0.773	0.880
费氏弧菌（Vibrio fischeri）	RMSE	0.739	0.724	0.658	0.732	0.696	0.591
	MAE	0.524	0.460	0.476	0.524	0.492	0.441
	R²	0.738	0.753	0.779	0.704	0.742	0.825

数据集	评估指标	SVM	RF	MLP	GAT	MPNN	GCM
大鼠白血病细胞（IPC-81）	RMSE	0.687	0.655	0.640	0.519	0.486	0.457
	MAE	0.510	0.436	0.467	0.379	0.353	0.325
	R²	0.683	0.706	0.688	0.764	0.787	0.867
对乙酰胆碱酯酶（AChE）	RMSE	0.229	0.286	0.243	0.295	0.246	0.229
	MAE	0.128	0.170	0.156	0.210	0.171	0.184
	R²	0.814	0.746	0.825	0.716	0.822	0.830
大肠杆菌（E. coli）	RMSE	0.545	0.547	0.665	0.648	0.610	0.396
	MAE	0.396	0.407	0.462	0.471	0.417	0.324
	R²	0.832	0.811	0.775	0.760	0.773	0.880
费氏弧菌（Vibrio fischeri）	RMSE	0.739	0.724	0.658	0.732	0.696	0.591
	MAE	0.524	0.460	0.476	0.524	0.492	0.441
	R²	0.738	0.753	0.779	0.704	0.742	0.825

序号	原子序号	原子基团	毒性贡献值W_atom
1	0	CH₃	0.437
2	12	F	-0.320
3	13	F	-0.320
4	14	F	-0.320
5	15	F	-0.320
6	10	F	-0.320
7	8	CH	0.440
8	7	CH	0.561
9	9	CH	0.622
10	11	P^-	0.413
11	1	CH₂	-0.017
12	16	F	-0.320
13	2	CH₂	0.605
14	3	CH₂	0.643
15	4	N⁺	1.058
16	5	CH	0.622
17	6	CH	0.440

序号	原子序号	原子基团	毒性贡献值W_atom
1	0	CH₃	0.437
2	12	F	-0.320
3	13	F	-0.320
4	14	F	-0.320
5	15	F	-0.320
6	10	F	-0.320
7	8	CH	0.440
8	7	CH	0.561
9	9	CH	0.622
10	11	P^-	0.413
11	1	CH₂	-0.017
12	16	F	-0.320
13	2	CH₂	0.605
14	3	CH₂	0.643
15	4	N⁺	1.058
16	5	CH	0.622
17	6	CH	0.440

数据集	正贡献原子基团	毒性贡献正权重	负贡献原子基团	毒性贡献负权重
大鼠白血病细胞（IPC-81）	S⁺	2.003	P^-	-0.960
	N⁺	0.975	C	-0.649
	NH⁺	0.937	CH₂	-0.562
	Cl^-	0.612	B^-	-0.545
	NH₂	0.595	P⁺	-0.432
对乙酰胆碱酯酶（AChE）	P⁺	1.985	F	-0.427
	N⁺	0.974	P	-0.366
	CH	0.659	C	-0.362
	C	0.604	S	-0.289
	S	0.589	N	-0.284
大肠杆菌（E. coli）	N⁺	1.293	B^-	-0.722
	S	0.858	P^-	-0.644
	CH₃	0.823	N⁺	-0.399
	P⁺	0.819	C	-0.399
	O^-	0.570	S	-0.385
费氏弧菌（Vibrio fischeri）	NH⁺	1.604	P^-	-1.206
	N^-	1.598	N	-0.930
	NH	1.306	S	-0.766
	N⁺	1.154	C	-0.759
	P	0.907	CH₂	-0.594