EVIDARA: Automated Evidential Support from Raw Data for relay agents in Biomedical KG Queries

EVIDARA：生物医学 KG 查询中中继代理的原始数据自动证据支持

• 批准号: 10547256
• 负责人: SERGIO E BARANZINI
• 金额: $ 53.29万
• 依托单位: INSTITUTE FOR SYSTEMS BIOLOGY
• 依托单位国家: 美国
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-01-24 至 2022-11-30
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10547256
• 关键词: Data; Text

1) Component: Autonomous Relay Agent. We will develop an ARA named EVIDARA to evaluate returns from queries in knowledge sources (KS) using a new epistemology: The “reasoning” is based on checking against empirical evidence available in raw data (measurements) instead of deductive reasoning (FIG.►). EVIDARA will assist the Autonomous Relay System (ARS) to identify paths in returned knowledge graphs (KG) that may conflict with real-word evidence and to relay queries to appropriate specialty KS or database. (2) Problem addressed: EHR and multi-omics raw data from large cohorts, if properly preprocessed [e.g., by Knowledge Providers, such as the DOCKET, see application by Dr. Glusman], offers a new opportunity for ad hoc systematic extraction of empirical knowledge on relationships (“Protein P level correlates with risk for disease D”) instead of relying on specific epidemiological analyses. The problem in harnessing raw data for empirical support in lieu of deductive reasoning is that the KGs to be evaluated are extracted from knowledge sources of distinct types and that the relevance of paths depends on the query context Q. Also the ARA algorithm should be scalable to digest the emerging multi-omics data from projects like All-of-Us, the UK Biobank. (3) Plan for implementation: Research will be conducted to evaluate a new epistemic realm: make empirical evidence central to “reasoning”. We have assembled a set of functioning tools to overcome the chicken-egg problem of getting a project started and jumpstart development and testing of EVIDARA: (i) SPOKE, one of the largest biomedical knowledge network (KN) has integrated 25 diverse of KS into a single (neo4j) network database of 2 million nodes and will serve as testing ground for research well before we can use KGs produced by the Knowledge Providers. (ii) Algorithms that use raw data from EHR and multi-omics studies to evaluate the returned KGs. For instance, we compute weights of all nodes in the entire KN through a random-walk algorithm biased by their role for a given condition Q observed in the raw data. (iii) Raw data beyond EHR: multi-omics profiles from a study at ISB with >10k variables which vastly exceeds coverage of observable nodes in KNs offered by EHRs. Example query: “Vitamin K stimulates stem-cell signaling, thus could promote cancer. What is the molecular pathway? Mechanisms returned as KG will be pruned by EVIDARA and checked against correlative evidence in the raw data: Is there evidence that taking Vit. K or its antagonist reduces cancer risk?”. Importantly, since EVIDARA learns on a network of many types of KS, it will provide information to the ARS about which type of KS/Knowledge Provider to invoke next (in iterative queries) to improve the knowledge graph. (4) Expertise & resources: The MPIs, Drs. S. Baranzini (UCSF) and S. Huang (ISB) are researchers with long history of working with medical big data, thus offering technical expertise and the critical SME perspective. SB’s team has created and maintains SPOKE. The uniquely self-contained SPOKE network will allow NCATS staff to test other ARAs. SH brings decades of experience in research of disease mechanisms and medical epistemology. His team will provide multi-omics datasets and data analytics expertise. With his prior work in the NCATS Translator program, he is well poised to maximize team science efficiency and help convert its vision into tangible results. (5) Potential challenges. (i) Quality of evidential support depends on quality of raw data. A quality control is beyond the scope of EVIDARA but could be provided by Knowledge Providers focusing on new multi-omics data sets (e.g. DOCKET). (ii) Testing EVIDARA on other KS from Knowledge Providers) may be slowed down by interoperability issues (e.g. incompatible identifiers). Such issues will be addressed early in Year 1 with help of the Standard and Reference group.

1)组件：自治中继代理。 我们将开发一个名为EVIDARA的ARA， 在知识中评估查询的返回值 来源（KS）使用新的认识论： “推理”是基于对经验的检验 原始数据中的可用证据（测量） 而不是演绎推理 (FIG.）。EVIDARA将协助自治 中继系统（ARS），以识别返回路径 知识图谱（KG）， 与真实证据冲突，并将查询中继到适当的专业KS或数据库。 (2)解决的问题：来自大型队列的EHR和多组学原始数据（如果经过适当预处理） [e.g.,由知识提供者，如DOCKET，参见Glusman博士的申请]， 提供了一个新的机会，专门系统地提取经验知识的关系， （“蛋白质P水平与疾病D的风险相关”），而不是依赖于特定的流行病学 分析。利用原始数据作为经验支持而不是演绎推理的问题 要评估的知识库是从不同类型的知识源中提取的， 路径的相关性取决于查询上下文Q。此外，ARA算法应该是可扩展的 消化来自英国生物银行All-of-Us等项目的新兴多组学数据。 (3)执行计划：将开展研究，以评估一个新的认识领域： 使经验证据成为“推理”的核心。我们已经组装了一套功能工具， 克服了先有鸡还是先有蛋的问题，启动了一个项目， EVIDARA的测试：（一）SPOKE，最大的生物医学知识网络（KN）之一， 25个不同的KS到一个单一的（neo4j）网络数据库的200万个节点，并将服务于 在我们使用知识提供者制作的幼儿园之前，我们必须先把它作为研究的试验场。 (ii)使用来自EHR和多组学研究的原始数据来评估返回的KG的算法。 例如，我们通过随机行走算法计算整个KN中所有节点的权重 由于它们在原始数据中观察到的给定条件Q的作用而有偏差。(iii)EHR以外的原始数据： 来自ISB的一项研究的多组学特征，具有> 10 k个变量，大大超过了 由EHR提供的KN中的可观察节点。示例查询：“维生素K刺激干细胞信号传导， 从而可能促进癌症。什么是分子途径？退回的机制为KG 将由EVIDARA进行修剪，并根据原始数据中的相关证据进行检查： 服用维生素的证据。K或其拮抗剂降低癌症风险？重要的是，由于EVIDARA 在许多类型的KS网络上学习，它将向ARS提供关于哪种类型的信息 KS/知识提供者的下一个调用（在迭代查询中）以改进知识图。 (4)专业知识和资源：MPI，S。Baranzini（UCSF）和S. Huang（ISB）是研究人员 拥有悠久的医疗大数据工作历史，从而提供技术专业知识和 关键的中小企业视角。SB的团队创建并维护了SPOKE。独一无二的自给自足 SPOKE网络将允许NCATS工作人员测试其他ARA。SH带来了数十年的经验 在疾病机制和医学认识论研究中的重要作用。他的团队将提供多组学 数据集和数据分析专业知识。凭借他在NCATS翻译程序之前的工作，他是 能够最大限度地提高团队科学效率，并帮助将其愿景转化为切实的成果。 (5)潜在的挑战。(i)证据支持的质量取决于原始数据的质量。质量 控制超出了EVIDARA的范围，但可以由知识提供者提供， 新的多组学数据集（例如DOCKET）。(ii)在Knowledge的其他KS上测试EVIDARA 互操作性问题（例如，不兼容的标识符）可能会减慢供应商的速度。等 这些问题将在标准和参考小组的帮助下在第一年早些时候得到解决。