知识图谱:一种系统性构建因果图的方法

doi:10.3969/j.issn.1006-7795.2022.04.012

首都医科大学学报 ›› 2022, Vol. 43 ›› Issue (4): 584-591.doi: 10.3969/j.issn.1006-7795.2022.04.012

• 医学信息学:应用和发展 • 上一篇下一篇

知识图谱:一种系统性构建因果图的方法

白永梅^1,2,3, 孙华鸽⁴, 杜建^2*

1.北京大学医学部医学技术研究院,北京 100191;
2.北京大学健康医疗大数据国家研究院,北京 100191;
3.北京大学医学部公共卫生学院,北京 100191;
4.墨尔本大学数学与统计学院,澳大利亚墨尔本 3010

收稿日期:2022-02-24 出版日期:2022-08-21 发布日期:2022-10-28
基金资助:
国家自然科学基金面上项目(72074006),中国科协青年人才托举工程项目(2017QNRC001),北京大学医学部人才启动经费项目(BMU2021YJ008).

A systematic approach to constructing DAG from knowledge graph

Bai Yongmei^1,2,3, Sun Huage⁴, Du Jian^2*

1. Institute of Medical Technology, Peking University Health Science Center, Peking University, Beijing 100191, China;
2. National Institute of Health Data Science, Peking University, Beijing 100191, China;
3. School of Public Health, Peking University, Beijing 100191, China;
4. School of Mathematics and Statistics, The University of Melbourne, Melbourne 3010, Australia

Received:2022-02-24 Online:2022-08-21 Published:2022-10-28
Contact: *E-mail:dujian@bjmu.edu.cn
Supported by:
This study was supported by National Natural Science Foundation of China (72074006), Young Elite Scientists Sponsorship Program by China Association for Science and Technology (2017QNRC001), Peking University Health Science Center(BMU2021YJ008).

摘要/Abstract

摘要： 因果推断(相对于相关性分析)是基于大数据的观察性研究的主要目标。因果图通过有向无环图(directed acyclic graph, DAG)整合大量先验知识将变量之间复杂的因果关系可视化,已成为合理制定因果推断策略的重要工具。然而目前因果图的构建主要依赖专家知识和经验,亟需从整个医学知识体系的角度进行系统化构建,从现有出版物中进行医学知识提取是系统构建DAG的基础,本文将系统介绍基于美国国立卫生研究院SemMedDB数据库研发的结构化医学知识体系平台。本文尝试从跨学科角度,将因果图定义为研究问题涉及概念(头概念和尾概念)及其所有第三方变量之间的复杂网络,提出系统化构建DAG提供新策略:一是将知识图谱修剪为因果图;二是将基于人群-干预/暴露-对照-结果(population-interventions/exposure-comparisons-outcomes,PI/ECO)框架的证据结论合成为因果图。

关键词: 因果图, 知识图谱, 证据合成, 混杂变量, 中间变量, 对撞变量

Abstract: Causal inference is the primary goal of observational research based on big data, as opposed to correlation analysis. Causal graph often visualize complex causal relationships by integrating large amount of priori knowledge through directed acyclic graph(DAG). The directed acyclic graph has become an important tool for developing causal inference strategies. However, the construction of causal graph for specific research questions currently relies heavily on expert knowledge and local experience. Medical knowledge extraction from existing publications is the basis for the systematic construction of DAG. In this paper, we will systematically introduce the structured medical knowledge system platform developed based on the SemMedDB database of the National Institutes of Health. This study attempts to provide a new strategy for systematically constructing DAG by defining causal graphs from an interdisciplinary perspective as complex networks between the concepts involved in a research problem (head and tail concepts) and all their third-party variables. There are two main approaches for the current systematic construction of causal graphs: (1) Prune the knowledge graph into causal graph; (2) Combine the evidence claims based on population-interventions/exposure-comparisons-outcomes (PI/ECO) framework into causal graphs.

Key words: directed acyclic graph, knowledge graph, evidence synthesis, confounder, mediator, collider

中图分类号:

白永梅, 孙华鸽, 杜建. 知识图谱:一种系统性构建因果图的方法[J]. 首都医科大学学报, 2022, 43(4): 584-591.

Bai Yongmei, Sun Huage, Du Jian. A systematic approach to constructing DAG from knowledge graph[J]. Journal of Capital Medical University, 2022, 43(4): 584-591.

参考文献

[1] Zuccolo L, Holmes M V. Commentary: mendelian randomization-inspired causal inference in the absence of genetic data[J]. Int J Epidemiol, 2017, 46(3):962-965.
[2] 郑英杰, 赵耐青, 何一宁. 客观世界的因果关系: 基于有向无环图的结构解析[J]. 中华流行病学杂志, 2018, 39(1):90-93.
[3] Barnard-Mayers R, Kouser H, Cohen J A, et al. A case study and proposal for publishing directed acyclic graphs: the effectiveness of the quadrivalent human papillomavirus vaccine in perinatally HIV Infected girls[J]. J Clin Epidemiol, 2022, 144: 127-135.
[4] Bareinboim E, Pearl J. Causal inference and the data-fusion problem[J]. Proc Natl Acad Sci U S A, 2016, 113(27):7345-7352.
[5] Ioannidis J P A, Tan Y J, Blum M R. Limitations and misinterpretations of E-values for sensitivity analyses of observational studies[J]. Ann Intern Med, 2019, 170(2):108-111.
[6] Pearl J. Causal inference from indirect experiments[J]. Artif Intell Med, 1995, 7(6):561-582.
[7] Greenland S, Pearl J, Robins J M. Causal diagrams for epidemiologic research[J]. Epidemiology, 1999, 10(1):37-48.
[8] Howards P P, Schisterman E F, Poole C, et al. “Toward a clearer definition of confounding” revisited with directed acyclic graphs[J]. Am J Epidemiol, 2012, 176(6):506-511.
[9] Prada-Ramallal G, Takkouche B, Figueiras A. Bias in pharmacoepidemiologic studies using secondary health care databases: a scoping review[J]. BMC Med Res Methodol, 2019, 19(1):53.
[10] 荆典, 段重阳, 谭铭, 等. 基于有向无环图和线性回归理论定量分析常见混杂的影响[J]. 中国卫生统计, 2021, 38(4):508-513, 518.
[11] Malec S A, Wei P, Bernstam E V, et al. Using computable knowledge mined from the literature to elucidate confounders for EHR-based pharmacovigilance[J]. J Biomed Inform, 2021, 117: 103719.
[12] Angrist J D, Imbens G W, Rubin D B. Identification of causal effects using instrumental variables[J]. J Am Stat Assoc, 1996, 91(434):444-455.
[13] Swanson D R, Smalheiser N R. An interactive system for finding complementary literatures: a stimulus to scientific discovery[J]. Artif Intell, 1997, 91(2):183-203.
[14] Swanson D R. Fish oil, Raynaud's syndrome, and undiscovered public knowledge[J]. Perspect Biol Med, 1986, 30(1):7-18.
[15] Weeber M, Klein H, De Jong-Van Den Berg L T W, et al. Using concepts in literature-based discovery: simulating Swanson's Raynaud-fish oil and migraine-magnesium discoveries[J]. J Am Soc Inf Sci Technol, 2001, 52(7):548-557.
[16] Nordon G, Koren G, Shalev V, et al. Building causal graphs from medical literature and electronic medical records[C]//Proceedings of the Thirty-Third AAAI Conference on Artificial Intelligence and Thirty-First Innovative Applications of Artificial Intelligence Conference and Ninth AAAI Symposium on Educational Advances in Artificial Intelligence.Honolulu, Hawaii, USA: AAAI Press, 2019: 1102-1109.
[17] Nordon G, Koren G, Shalev V, et al. Separating wheat from chaff: joining biomedical knowledge and patient data for repurposing medications[C]//Proceedings of the Thirty-Third AAAI Conference on Artificial Intelligence and Thirty-First Innovative Applications of Artificial Intelligence Conference and Ninth AAAI Symposium on Educational Advances in Artificial Intelligence, Honolulu, Hawaii, USA: AAAI Press, 2019: 9565-9572.
[18] Rindflesch T C, Fiszman M. The interaction of domain knowledge and linguistic structure in natural language processing: interpreting hypernymic propositions in biomedical text[J]. J Biomed Inform, 2003, 36(6):462-477.
[19] Gyori B M, Bachman J A, Subramanian K, et al. From word models to executable models of signaling networks using automated assembly[J]. Mol Syst Biol, 2017, 13(11):954.
[20] 唐丹, 肖雄, 杨帆, 等. 因果图模型及其在营养流行病学研究中的应用[J]. 中华流行病学杂志, 2021, 42(10):1882-1888.
[21] Ferguson K D, McCann M, Katikireddi S V, et al. Evidence synthesis for constructing directed acyclic graphs (ESC-DAGs):a novel and systematic method for building directed acyclic graphs[J]. Int J Epidemiol, 2020, 49(1):322-329.
[22] Caliendo M, Kopeinig S. Some practical guidance for the implementation of propensity score matching[J]. J Econ Surv, 2008, 22(1):31-72.
[23] Imbens G W, Angrist J D. Identification and estimation of local average treatment effects[J]. Econometrica, 1994, 62(2):467-475.
[24] Robins J M, Hernán M A, Brumback B. Marginal structural models and causal inference in epidemiology[J]. Epidemiology, 2000, 11(5):550-560.
[25] Kuang K, Li Y Z, Li B, et al. Continuous treatment effect estimation via generative adversarial de-confounding[J]. Data Min Knowl Discov, 2021, 35(6):2467-2497.
[26] Liu Y, Elsworth B, Erola P, et al. EpiGraphDB: a database and data mining platform for health data science[J]. Bioinformatics, 2021, 37(9):1304-1311.
[27] Kilicoglu H, Shin D, Fiszman M, et al. SemMedDB: a PubMed-scale repository of biomedical semantic predications[J]. Bioinformatics, 2012, 28(23):3158-3160.
[28] Röhrig N, Strobl R, Müller M, et al. Directed acyclic graphs helped to identify confounding in the association of disability and electrocardiographic findings: results from the KORA-Age study[J]. J Clin Epidemiol, 2014, 67(2):199-206.
[29] Tennant P W G, Murray E J, Arnold K F, et al. Use of directed acyclic graphs (DAGs) to identify confounders in applied health research: review and recommendations[J]. Int J Epidemiol, 2021, 50(2):620-632.
[30] Riseberg E, Melamed R D, James K A, et al. Development and application of an evidence-based directed acyclic graph to evaluate the associations between metal mixtures and cardiometabolic outcomes[DB/OL].(2021-03-08)[2022-02-24].https://doi.org/10.1101/2021.03.05.21252993.
[31] Henry S, McInnes B T. Literature based discovery: models, methods, and trends[J]. J Biomed Inform, 2017, 74: 20-32.
[32] Kastrin A, Hristovski D. Who is who in literature-based discovery: preliminary analysis[M]//Lu W, Zhu K Q. Trends and applications in knowledge discovery and data mining. Cham:Springer International Publishing, 2020: 51-59.

知识图谱:一种系统性构建因果图的方法

A systematic approach to constructing DAG from knowledge graph

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 1

编辑推荐

Metrics

本文评价