-
系统毒理学是建立在系统生物学基础上,综合多组学分析和传统毒理学方法,借助生物信息学[1-3]和计算毒理学[4-6]等模型化信息整合技术,对生物系统在外源化学物质扰动下保持稳定的能力进行评估,研究外源化学物质与生物系统相互作用机制的一门学科[7-9]。在还原论[10]思想引导下的毒理学,为研究某一种物质的毒性效应,只需要寻找并鉴别出与这一毒性效应直接相关的分子靶点即可。这是一种从整体到局部的简化研究思路。然而生物体信号通路的输入输出并不是由单一靶点控制的,而是由该通路的系统性结构和动力学共同控制的。有别于还原论思想,系统毒理学的核心思想是细胞和有机体等个体水平的形态和功能变化是由基因组、转录组和蛋白质组等系统扰动共同引起的。要理解化学物质在系统层次造成的生物学影响,就必须研究细胞和有机体整体的结构、功能特性和动力学机制,而非孤立部分的特征。系统毒理学分析框架强调系统性,可以减少跨物种外推、高剂量外推和低剂量外推的不确定性,有助于理解外源化合物对生命不同阶段的毒理作用以及遗传因素等对毒理作用的影响。
自21世纪初,组学技术为系统层次的计算建模奠定了数据基础[11]。从系统角度理解外源化学物质对生物的系统毒理学影响方面的研究逐渐增加[12-14]。如今多种组学技术的进步与组合,包括高精度的分子测量手段、高通量和高内涵的表征方法以及不断增强的计算能力、数据存储能力和信息管理工具,赋予了系统毒理学新的发展动力[15-17]。这些技术提供了海量的数据,也促使高质量数据库的建立[18].
组学数据可分为系统生物学数据和暴露相关数据两类,具有多维度、多尺度、彼此关联等特征。多维度是指整合的多组学数据描述了生物系统基因、蛋白质等不同维度的分子机制。多尺度是指组学数据既可表征生物系统的整体结构功能,又可用于描述局部的代谢通路。多组学数据彼此的关联性,体现在单一维度的组学研究无法解释某些生物表型现象[19-20]。
面对这些这些数据带来的发展机遇,美国国家研究咨询委员会编写了“21世纪毒性测试:远景与策略”这一研究报告。报告指出系统模型被期望建立以“毒性通路”为核心的量效关系模型,并基于给定的暴露条件将in vitro结果外推到in vivo人体血液和组织浓度[21]。为实现该目标,模型需要具备精准识别化学物质对生物系统扰动的能力。任何生物系统都由相互作用、相互依赖的若干组成部分结合而成,是具有特定功能的有机整体。这样的性质,可引导寻找能表示系统中各组分之间的相互作用关系的数学工具[22-23]。基于网络的模型有通用性强、灵活性强、包含节点间关系信息等优势,在系统毒理学中起到整合多组学数据、挖掘不同尺度的生物学信息的关键作用,得到了广泛应用[24-27]。
网络在数学上亦称图(Graph)[28]。基因调控和蛋白质相互作用等组学数据均可用网络表征。网络是节点与边的集合,网络中节点、边和网络整体结构3种信息可以共同用于描绘生物系统的结构功能。然而也正是这种灵活但不规则的数据结构对数据的处理和建模造成了很大阻碍。因此,以往系统毒理学对网络模型的分析方法局限于网络基本的拓扑学性质指标,例如节点的度(该节点连接的其他节点的数目)、节点的聚类系数(与该节点相邻的所有节点之间连边的数目占这些相邻节点之间最大可能连边数目的比例)等。已有较成熟的软件可以对网络图进行可视化与统计分析,如Cytoscape[29]、Pajek[30]。这种研究方式在慢性疾病防控、化学品风险评价等领域被应用[31-34]。然而,在人为选择用于描述网络的拓扑学性质指标时,网络中的部分信息被直接忽略,造成了信息损失。
近年来,信息科学领域开展了诸多针对网络的算法研究,可分为网络嵌入[35-36]与图神经网络(GNN)[37]。网络嵌入是一种将网络的结构特征整合进节点特征,进而实现网络特征化表示的信息挖掘手段。GNN则可以对大规模网络中的信息详尽地提取和学习[38],在基于网络的系统毒理学研究中有很大的应用潜力。其中,比较有代表性的算法是图卷积网络算法(GCN)[39]。GCN是一种深度学习算法,其优势在于能够从大规模网络数据中自动学习输入特征和输出决策之间的复杂关系,从而实现“端到端”的学习。换言之,在深度学习的特征提取过程中,模型不需要人为选择描述网络的拓扑指标作为模型的输入特征,而是可直接将用于描述生物系统的网络作为输入。基于此建立的模型,可充分利用网络中节点、边和网络结构3种信息,根据实际任务需求,综合完成如节点的分类预测、边的连接预测和网络图整体的分类预测等不同种类的预测任务。
多组学数据的复杂性,使得以前的大多数研究仅仅侧重于分析单一类型的组学数据。组学间的相互作用尚未得到充分探讨[40]。基于网络的模型,可从多组学数据中提取数据间的逻辑关系。但传统网络分析方法在异质网络上的信息提取能力和计算效率较差[41]。因此,GNN模型有望被用于构建多维度、多尺度的先进系统毒理学模型,用于化学品有害效应的模拟预测(图1)。
本文介绍网络分析方法的研究策略、在系统毒理学领域中的应用,并对GNN模型在系统毒理学领域的应用进行展望。
基于图神经网络模型的系统毒理学研究展望
Prospect of systems toxicology research based on graph neural network model
-
摘要: 系统毒理学是建立在系统生物学基础上,综合多组学分析和传统毒理学方法,借助生物信息学和计算毒理学等模型化信息整合技术,对生物系统在外源化学物质扰动下保持稳定的能力进行评估,研究外源化学物质与生物系统相互作用机制的一门学科。转录组、蛋白质组、代谢组、暴露组等多组学数据,有多维度、多尺度、多关联的特征,为系统毒理学建模奠定了数据基础。如何利用计算建模,对多组学数据进行有效挖掘成为有待攻克的瓶颈。针对多组学数据的特点,基于网络的模型有着通用性强、灵活性强、包含节点间关系信息等优势,在系统毒理学中起到整合与挖掘多组学数据的关键作用。图神经网络(GNN)作为一种深度学习方法,在系统毒理学建模中展现了良好的应用前景。本文介绍了系统毒理学的研究目的、网络分析方法的研究策略,对GNN在系统毒理学领域的应用进行了展望。
-
关键词:
- 系统毒理学 /
- 多组学 /
- 网络分析方法 /
- 图神经网络(GNN)
Abstract: Systems toxicology can be regarded as an interdiscipline that is based on systems biology and combines multi-omics data analysis and classical toxicology methods. With information integration technologies such as bioinformatics and computational toxicology, systems toxicology can evaluate ability of biological systems to maintain stability under perturbation of xenobiotics and describe interactions of xenobiotics with biological systems. Transcriptomics, proteomics, metabolomics and exposomics data are multi-dimensional, multi-scaled and multi-correlational, and can lay a data foundation for systems toxicology modeling. It has become a technological bottleneck to use computational modeling for effective mining the multi-omics data. Network-based models exhibit advantages of high generality and flexibility, can characterize relationship information between nodes, and thus are suitable for dealing with multi-omics data, and can play a key role in integrating and mining multi-omics data in systems toxicology. Graph neural network (GNN) as a deep learning method, shows good application prospects in systems toxicology modeling. In this perspective, research orientations of systems toxicology and research strategies of network analysis methods are introduced, and application of GNN in systems toxicology is envisaged.-
Key words:
- systems toxicology /
- multi-omics /
- network analysis method /
- graph neural network (GNN)
-
[1] HAGEN J B. The origins of bioinformatics [J]. Nature Reviews Genetics, 2020, 1: 231-236. [2] KANEHISA M, BORK P. Bioinformatics in the post-sequence era [J]. Nature Genetics, 2003, 33: 305-310. doi: 10.1038/ng1109 [3] GAUTHIER J, VINCENT A T, CHARETTE S J, et al. A brief history of bioinformatics [J]. Briefings in Bioinformatics, 2019, 20(6): 1981-1996. doi: 10.1093/bib/bby063 [4] 陈景文, 王中钰, 傅志强. 环境计算化学与毒理学[M]. 北京: 科学出版社, 2018, 22-34. CHEN J W, WANG Z Y, FU Z Q. Computational chemistry and toxicology of the environment[M]. Beijing: Science Press, 2018, 22-34(in Chinese).
[5] 王中钰, 陈景文, 乔显亮, 等. 面向化学品风险评价的计算(预测)毒理学 [J]. 中国科学:化学, 2016, 46(2): 222-240. WANG Z Y, CHEN J W, QIAO X L, et al. Computational toxicology: oriented for chemicals risk assessment [J]. Science China Chemistry, 2016, 46(2): 222-240(in Chinese).
[6] KAVLOCK R, DIX D. Computational toxicology as implemented by the US EPA: Providing high throughput decision support tools for screening and assessment chemical exposure, hazard and risk [J]. Journal of Toxicology and Environmental Health, Part B:Critical Reviews, 2010, 13(2-4): 197-217. doi: 10.1080/10937404.2010.483935 [7] 王先良, 徐顺清. 系统毒理学及其应用 [J]. 生态毒理学报, 2006, 1(4): 289-294. WANG X L, XU S Q. Systems toxicology [J]. Asian Journal of Ecotoxicology, 2006, 1(4): 289-294(in Chinese).
[8] SHANA J S, ALAN R B, REX E F, et al. Systems toxicology: from basic research to risk assessment [J]. Chemical Research in Toxicology, 2014, 27(3): 314-329. doi: 10.1021/tx400410s [9] THOMAS H, REX E F, PAUL J, et al. Systems toxicology: real world applications and opportunities [J]. Chemical Research in Toxicology, 2017, 30(4): 870-882. doi: 10.1021/acs.chemrestox.7b00003 [10] 桂起权. 解读系统生物学: 还原论与整体论的综合 [J]. 自然辩证法通讯, 2015, 37(5): 219. GUI Q Q. Reading of systems biology: integration of reductionism and holism [J]. Journal of Dialectics of Nature, 2015, 37(5): 219(in Chinese).
[11] AARDEMA M J, MACGREGOR J T. Toxicology and genetic toxicology in the new era of 'Toxicogenomics': impact of '-Omics' technologies [J]. Mutation Research, 2002, 499(1): 13-25. doi: 10.1016/S0027-5107(01)00292-5 [12] IDEKER T, GALITSKI T, HOOD L. A new approach to decoding life: systems biology [J]. Annual Review of Genomics and Human Genetics, 2001, 2: 343-372. doi: 10.1146/annurev.genom.2.1.343 [13] WATERS M D, BOORMAN G, BUSHEL P, et al. Systems toxicology and the chemical effects in biological systems (CEBS) knowledge base [J]. Environmental Health Perspectives, 2003, 111(6): 811-824. doi: 10.1289/ehp.5971 [14] KITANO H. Systems biology: A brief overview [J]. Science, 2002, 295(5560): 1662-1664. doi: 10.1126/science.1069492 [15] WATERS M D, FOSTEL J M. Toxicogenomics and systems toxicology: aims and prospects [J]. Nature Reviews Genetics, 2004, 5(12): 936-948. doi: 10.1038/nrg1493 [16] HARTUNG T, VLIET E, JAWORSKA J, et al. Food for thought [J]. Systems Toxicology ALTEX, 2012, 29: 119-128. [17] PLANT N J. An introduction to systems toxicology [J]. Toxicology Research, 2015, 4: 9-22. doi: 10.1039/C4TX00058G [18] 李杰, 李柯佳, 张臣, 等. 计算系统毒理学: 形成、发展及应用 [J]. 科学通报, 2015, 60(19): 1751-1760. doi: 10.1360/N972014-01400 LI J, LI K J, ZHANG C, et al. Computational systems toxicology: Emergence, development and application [J]. Chinese Science Bulletin, 2015, 60(19): 1751-1760(in Chinese). doi: 10.1360/N972014-01400
[19] QUINN R A, MELNIK A V, VRBANAC A, et al. Global chemical effects of the microbiome include new bile-acid conjugations [J]. Nature, 2020, 579: 123-129. doi: 10.1038/s41586-020-2047-9 [20] KIMURA I, MIYAMOTO J, KITANO R O, et al. Maternal gut microbiota in pregnancy influences offspring metabolic phenotype in mice [J]. Science, 2020, 367(6481): 8429. doi: 10.1126/science.aaw8429 [21] KREWSKI D, JR. A D, ANDERSEN M, et al. Toxicity testing in the 21st century: A vision and a strategy [J]. Journal Toxicology Environmental Health-Part B-Critical Reviews, 2010, 13: 51-138. doi: 10.1080/10937404.2010.483176 [22] ZHANG Q, BHATTACHARYA S, ANDERSEN M E, et al. Computational systems biology and dose-response modeling in relation to new directions in toxicity testing [J]. Journal of Toxicology and Environmental Health, Part B:Critical Reviews, 2010, 13(2-4): 253-276. doi: 10.1080/10937404.2010.483943 [23] ZHANG Q, BHATTACHARYA S, CONOLLY R B, et al. Molecular signaling network motifs provide a mechanistic basis for cellular threshold responses [J]. Environmental Health Perspectives, 2014, 122(12): 61-70. [24] SHAO Z M, WANG K K, ZHANG S Y, et al. Ingenuity pathway analysis of differentially expressed genes involved in signaling pathways and molecular networks in RhoE gene-edited cardiomyocytes [J]. International Journal of Molecular Medicine, 2020, 46(3): 1225-1238. doi: 10.3892/ijmm.2020.4661 [25] KANEHISA M, GOTO S, FURUMICHI M, et al. KEGG for representation and analysis of molecular networks involving diseases and drugs [J]. Nucleic Acids Research, 2010, 28(Suppl_1): 355-360. [26] ZHANG Y J, LIN H F, YANG Z H, et al. A method for predicting protein complex in dynamic PPI networks [J]. BMC Bioinformatics, 2016, 17(7): 229. [27] MANIPUR I, GRANATA I, MADDALENA L, et al. Clustering analysis of tumor metabolic networks [J]. BMC Bioinformatics, 2020, 21: 349. doi: 10.1186/s12859-020-03564-9 [28] BATTAGLIA P W, HAMRICK J B, BAPST V, et al. Relational inductive biases, deep learning, and graph networks [EB/OL]. [2021-5-15]. arXiv preprint, 2018,https://export.arxiv.org/pdf/1806.01261. [29] SHANNON P, MARKIEL A, OZIER O, et al. Cytoscape: A software environment for integrated models of biomolecular interaction networks [J]. Genome Research, 2003, 13: 2498-2504. doi: 10.1101/gr.1239303 [30] BATAGELJ V, MRVAR A. Pajek-program for large network analysis [J]. Connections, 1998, 21: 47-57. [31] JAN S, ANNE C G. SnapShot: protein-protein interaction networks [J]. Cell, 2011, 144(6): 1000-1001. doi: 10.1016/j.cell.2011.02.025 [32] JOERG M, AMITABH S, MAKSIM K, et al. Uncovering disease-disease relationships through the incomplete interactome [J]. Science, 2015, 347(6224): 1257601. doi: 10.1126/science.1257601 [33] KARINE A, PHILIPPE G. Application of computational systems biology to explore environmental toxicity hazards [J]. Environmental Health Perspectives, 2011, 119(12): 1754-1759. doi: 10.1289/ehp.1103533 [34] DAI W N, TANG T T, DAI Z H, et al. Probing the mechanism of hepatotoxicity of hexabromocyclododecanes through toxicological network analysis [J]. Environmental Science & Technology, 2020, 54(23): 15235-15245. [35] GROVER A, LESKOVEC J. Node2vec: scalable feature learning for networks [EB/OL]. [2021-5-15]. arXiv preprint, 2016, https://cs.stanford.edu/people/jure/pubs/node2vec-kdd16.pdf. [36] FIGUEIREDO D R, RIBEIRO L F R, SAVERESE P H P. Struc2vec: Learning node representations from structural identity [EB/OL]. [2021-5-15]. arXiv preprint, 2017, https://export.arxiv.org/pdf/1704.03165. [37] SEJNOWSKI T J. The unreasonable effectiveness of deep learning in artificial intelligence[J]. Proceedings of the National Academy of Sciences of the United States of America, 2020, 117(48): 30033-30038. [38] XU K, HU W, LESKOVEC J, et al. How powerful are graph neural networks [EB/OL]. [2021-5-15]. arXiv preprint, 2018, https://export.arxiv.org/pdf/1810.00826. [39] BRUNA J, ZAREMBA W, SZLAM A, et al. Spectral networks and locally connected networks on graphs [EB/OL]. [2021-5-15]. arXiv preprint, 2013,https://arxiv.org/pdf/1312.6203.pdf. [40] LIU Q, HU Z Q, JIANG R, et al. DeepCDR: A hybrid graph convolutional network for predicting cancer drug response [J]. Bioinformatics, 2020, 26(Supplement_2): 911-918. [41] ZHANG F, WANG M H, XI J N, et al. A novel heterogeneous network-based method for drug response prediction in cancer cell lines [J]. Scientific Reports, 2018, 8: 3355. doi: 10.1038/s41598-018-21622-4 [42] SCHAFF J, FINK C, SLEPCHENKO B, et al. A general computational framework for modeling cellular structure and function [J]. Biophysical Journal, 1997, 73(3): 1135-1146. doi: 10.1016/S0006-3495(97)78146-3 [43] YAHYA F A, HASHIM N F, ALI D A I, et al. A brief overview to systems biology in toxicology:The journey from in to vivo, in-vitro and -omics [J]. Journal of King Saud University-Science, 2020, 33(1): 101254. [44] KRISTIN S, BEAT B F, DANIELLE J M, et al. Transcriptomics in ecotoxicology [J]. Analytical and Bioanalytical Chemistry, 2010, 397(3): 917-923. doi: 10.1007/s00216-010-3662-3 [45] LANGFELDER P, HORVATH S. WGCNA: An package for weighted correlation network analysis [J]. BMC Bioinformatics, 2008, 9(559): 1471-2105. [46] TIAN Z L, HE W X, TANG J N, et al. Identification of important modules and biomarkers in breast cancer based on WGCNA [J]. OncoTargets and Therapy, 2020, 13: 6805-6817. doi: 10.2147/OTT.S258439 [47] 陈铭. 生物信息学(第3版)[M]. 北京: 科学出版社, 2018, 118-121. CHEN M. Bioinformatics (Third Edition) [M]. Beijing: Science Press, 2018, 118-121(in Chinese).
[48] KANEHISA M, FURUMICHI M, TANABE M, et al. KEGG: new perspectives on genomes, pathways, diseases and drugs [J]. Nucleic Acids Research, 2016, 45(D1): 353-361. [49] OBERHARDT M A, PUCHAIKA J, MARTINS V A P, et al. Reconciliation of genome-scale metabolic reconstructions for comparative systems analysis [J]. PLOS Computational Biology, 2011, 7(3): 1001116. doi: 10.1371/journal.pcbi.1001116 [50] PITKÄNEN E, JOUHTEN P, HOU J, et al. Comparative genome-scale reconstruction of gapless metabolic networks for present and ancestral species [J]. PLOS Computational Biology, 2014, 10(2): 1003465. doi: 10.1371/journal.pcbi.1003465 [51] KARLSEN E, SCHULZ C, ALMAAS E. Automated generation of genome-scale metabolic draft reconstructions based on KEGG [J]. BMC Bioinformatics, 2018, 19: 467. doi: 10.1186/s12859-018-2472-z [52] KRÄMER A, GREEN J, POLLARD J, et al. Causal analysis approaches in ingenuity pathway analysis [J]. Bioinformatics, 2014, 30(4): 523-530. doi: 10.1093/bioinformatics/btt703 [53] YUAN Y, JOSEPH Z B. GCNG: graph convolutional networks for inferring gene interaction from spatial transcriptomics data [J]. Genome Biology, 2020, 21: 200. doi: 10.1186/s13059-020-02088-y [54] WILKINS M R, SANCHEZ J C, GOOLEY A, et al. Progress with proteome projects: why all proteins expressed by a genome should be identified and how to do it [J]. Biotechnology and Genetic Engineering Reviews, 1996, 13(1): 19-50. doi: 10.1080/02648725.1996.10647923 [55] LEONIDAS G A, JULIO S R, BENJAMIN D C, et al. Networks inferred from biochemical data reveal profound differences in toll-like receptor and inflammatory signaling between normal and transformed hepatocytes [J]. Molecular & Cellular Proteomics, 2010, 9(9): 1849-1865. [56] BOLTZ T A, DEVKOTA P, WUCHTY S. Collective influencers in protein interaction networks [J]. Scientific Reports, 2019, 9: 3948. doi: 10.1038/s41598-019-40410-2 [57] RAMIREZ T, DANESHIAN M, KAMP H, et al. Metabolomics in toxicology and preclinical research [J]. ALTEX-Alternatives to Animal Experimentation, 2013, 30(2): 209-225. [58] LEONARDO D S, SALEH A, YARIV B, et al. Network-based strategies in metabolomics data analysis and interpretation: from molecular networking to biological interpretation [J]. Expert Review of Proteomics, 2020, 17(4): 243-255. doi: 10.1080/14789450.2020.1766975 [59] LI X K, YANG H J, XIAO J C, et al. Network pharmacology-based investigation into the bioactive compounds and molecular mechanisms of schisandrae chinensis fructus against drug-induced liver injury [J]. Bio-organic Chemistry, 2020, 96: 103553. [60] LI X K, LI M Y, DENG S, et al. A network pharmacology-integrated metabolomics strategy for clarifying the action mechanisms of schisandrae chinensis fructus for treating drug-induced liver injury by acetaminophen [J]. Bio-organic & Medicinal Chemistry, 2021, 31: 115992. [61] WILD P C. Complementing the genome with an "Exposome":The outstanding challenge of environmental exposure measurement in molecular epidemiology [J]. Cancer Epidemiology Biomarkers & Prevention, 2005, 14(8): 1847-1850. [62] CHAKRAVARTI A, LITTLE P. Nature, nurture and human disease [J]. Nature, 2003, 421: 412-414. doi: 10.1038/nature01401 [63] RAPPAPORT S M. Discovering environmental causes of disease [J]. Journal of Epidemiology and Community Health, 2012, 66(2): 99-102. doi: 10.1136/jech-2011-200726 [64] KALLOO G, WELLENIUS G A, MCCANDLESS L, et al. Profiles and predictors of environmental chemical mixture exposure among pregnant women: the health outcomes and measures of the environment study [J]. Environmental Science & Technology, 2018, 52(17): 10104-10113. [65] CHEN H, ZHANG W X, ZHOU Y Q, et al. Characteristics of exposure to multiple environmental chemicals among pregnant women in Wuhan, China [J]. Science of the Total Environment, 2021, 754: 142167. doi: 10.1016/j.scitotenv.2020.142167 [66] ROBINSON O, BASAGANA X, AGIER L, et al. The pregnancy exposome: multiple environmental exposures in the inma-sabadell birth cohort [J]. Environmental Science & Technology, 2015, 49(17): 10632-10641. [67] VINCENT B, RUTHANN A R. Mapping the human exposome to uncover the causes of breast cancer [J]. International Journal of Environmental Research and Public Health, 2020, 17(1): 189. [68] RUIZ C, ZITNIK M, LESKOVEC J. Identification of disease treatment mechanisms through the multiscale interactome [J]. Nature Communications, 2021, 12: 1796. doi: 10.1038/s41467-021-21770-8 [69] LIU X M, ENRICO M, ARDA H, et al. Robustness and lethality in multilayer biological molecular networks [J]. Nature Communications, 2020, 11: 6043. doi: 10.1038/s41467-020-19841-3 [70] STEAD W. Clinical implications and challenges of artificial intelligence and deep learning [J]. JAMA, 2018, 320(11): 1107-1108. doi: 10.1001/jama.2018.11029 [71] SUN M Y, ZHAO S D, GILVARY C, et al. Graph convolutional networks for computational drug development and discovery [J]. Briefings in Bioinformatics, 2019, 21(3): 919-935. [72] TANG W, CHEN J W, WANG Z Y, et al. Deep learning for predicting toxicity of chemicals: A mini review [J]. Journal of Environmental Science and Health Part C- Environmental Carcinogenesis and Ecotoxicology Reviews, 2018, 36(4): 252-271. [73] ZHANG Z, CUI P, ZHU W. Deep learning on graphs: A survey [EB/OL]. [2021-5-15]. arXiv Preprint, 2018, https://export.arxiv.org/pdf/1812.04202. [74] ZHOU J, CUI G, ZHANG Z. Graph neural networks: A review of methods and applications [EB/OL]. [2021-5-15]. arXiv preprint, 2019, https://export.arxiv.org/ftp/arxiv/papers/1812/1812.08434.pdf. [75] GILMER J, SCHOENHOLZ S, RILEY P F, et al. Neural message passing for quantum chemistry [EB/OL]. [2021-5-15]. arXiv preprint, 2017, https://www.ics.uci.edu/~mohamadt/papers/Neural_message_passing.pdf. [76] HAMILTON W L, YING R, LESKOVEC J. Inductive representation learning on large graphs [EB/OL]. [2021-5-15]. arXiv preprint, 2017, https://cs.stanford.edu/people/jure/pubs/graphsage-nips17.pdf. [77] ATWOOD J, TOWSLEY D. Diffusion-convolutional neural networks [EB/OL]. [2021-5-15]. arXiv preprint, 2015, https://papers.nips.cc/paper/2016/file/390e982518a50e280d8e2b535462ec1f-Paper.pdf. [78] NIEPERT M, AHMED M, KUTZKOV K. Learning convolutional neural networks for graphs [EB/OL]. [2021-5-15]. arXiv preprint, 2016, http://proceedings.mlr.press/v48/niepert16.pdf. [79] KRIZHEVSKY A, SUTSKEVER L, HINTON G E. ImageNet classification with deep convolutional neural networks[C]. In Proceedings of the 25th International Conference on Neural Information Processing Systems, 2012, 1: 1097-1105. [80] ALLEN T E H, GOODMAN J M, GUTSELL S, et al. Defining molecular initiating events in the adverse outcome pathway framework for risk assessment [J]. Chemical Research in Toxicology, 2014, 27(12): 2100-2112. doi: 10.1021/tx500345j [81] ANKLEY G T, BENNETT R S, ERICKSON R J, et al. Adverse Outcome Pathways: A conceptual framework to support ecotoxicology research and risk assessment [J]. Environmental Toxicology and Chemistry, 2010, 29(3): 730-741. doi: 10.1002/etc.34 [82] KREWSKI D, ANDERSEN M E, TYSHENKO M G, et al. Toxicity testing in the 21st century: Progress in the past decade and future perspectives [J]. Archives of Toxicology, 2020, 94(1): 1-58. doi: 10.1007/s00204-019-02613-4 [83] CHEN S, ZHANG Z H, QING T, et al. Activation of the Nrf2 signaling pathway in Usnic Acid-induced toxicity in HepG2 cells [J]. Archives of Toxicology, 2017, 91: 1293-1307. doi: 10.1007/s00204-016-1775-y [84] LIU L, WU F Y, ZHU C Y, et al. Involvement of dopamine signaling pathway in neurodevelopmental toxicity induced by isoniazid in zebrafish [J]. Chemosphere, 2021, 265: 129109. doi: 10.1016/j.chemosphere.2020.129109 [85] DREIER D A, DANIELLE F M, JOEL N M, et al. Linking mitochondrial dysfunction to organismal and population health in the context of environmental pollutants: Progress and considerations for mitochondrial adverse outcome pathways [J]. Environmental Toxicology and Chemistry, 2019, 38(8): 1625-1634. [86] SKARDING J, GABRYS B, MUSIAL K. Foundations and modelling of dynamic networks using dynamic graph neural networks: A survey [EB/OL]. [2021-5-15]. arXiv preprint, 2020, https://export.arxiv.org/pdf/2005.07496.