BRIDGE Center Standards Core

桥梁中心标准核心项目摘要 人工智能为发现新的生物医学见解提供了巨大的潜力，这些联系来自不同的、 跨域数据集。不幸的是，传统的假设驱动的数据集倾向于狭隘地关注 有针对性的问题领域，很少考虑“AI-Ready”。以最好地启用此类数据集的使用 在数据驱动和跨域发现中，必须使它们可查找、可访问、可互操作和 可重复使用(公平)。对于渴望数据的人工智能来说，缺乏公平性尤其是个问题。要充分利用 人工智能方法的力量，研究人员需要找到并重复使用数据来合并到更大的数据集，并且数据 必须是可互操作的或协调的，才能进行有意义的组合。将预先存在的数据集转换为 人工智能就绪数据具有挑战性，需要由人类专家进行广泛的链接和整理。这项挑战是 当跨域注释和链接数据时，这种情况会加剧，因为在这些域中，标准可能各不相同 和专一性。最后，许多数据集在数据透明度(包括内容)方面不遵守最佳做法 分配和再利用的归属和条件。这些关于可追溯性、许可、 和连接性为FIRE创造了一个可操作的模式：FIRE-TLC。 克服公平-TLC的障碍是翻译科学和人工智能驱动的生物医学发现的关键。我们的 团队领导了许多大型联盟的标准开发工作，包括GA4GH、HL7和 N3C。我们表示生物医学概念的标准已被广泛采用，包括 人类表型(如HPO、GA4GH表型)、疾病(如NCIT、MONDO、ICD-11)、基因(基因 本体论)、解剖学(Uberon)和分子变异(GA4GH VRS)。我们已经制定了标准和工具 解决数据来源(SEPIO)、贡献(贡献者归属模型)、许可障碍(数据 使用本体、可重用数据项目)和连接性(链接数据模型语言，LinkML)。 我们将在之前的工作、协作技能和技术知识的基础上开发一个框架，以 实现生物医学领域标准的统一。我们将成立工作组，与 数据生成项目(DGPS)的代表，用于记录用例和合成数据标准 要求。我们将提供指定标准的协议和培训，并提供礼宾服务 支持所有可交付成果和活动。我们将创建一个版本控制的Bridge2AI标准注册中心，以 DGPS使用的库存标准，在与通道无关的LinkML框架中指定，可发现 通过交互式标准中心，并通过我们的数据自动导出到技术构件 变换工具箱。我们将构建标准评估仪表板，用于评估和发现 Bridge2AI数据生成项目的数据集中的标准。我们将在 通过Duo、数据表和模型卡透明和负责任地共享数据集和ML模型。

BRIDGE Center Standards Core Project Summary AI offers great potential for the discovery of novel biomedical insights from linkages between disparate, cross-domain datasets. Unfortunately, traditional hypothesis-driven datasets tend to be narrowly focused on the targeted problem domain with little consideration to “AI-readiness”. To best enable the use of such datasets in data-driven and cross-domain discovery, they must be made Findable, Accessible, Interoperable, and Reusable (FAIR). Lack of FAIRness is particularly problematic for AI, which is data-hungry. To fully leverage the power of AI approaches, researchers need to find and reuse data to combine into larger datasets, and the data must be interoperable or harmonized to be combined meaningfully. Transforming pre-existing datasets into AI-ready data is challenging, requiring extensive linking and curation by human experts. This challenge is exacerbated when annotating and linking data across domains, where standards may be disparate in purpose and specificity. Finally, many datasets do not adhere to best practices in data transparency, including content attribution and conditions on distribution and reuse. These additional considerations of Traceability, Licensing, and Connectedness create an operationalized model for FAIR: FAIR-TLC. Overcoming the barriers to FAIR-TLC is key to translational science and AI-driven biomedical discovery. Our team has led standards development efforts in numerous large consortia, including the GA4GH, HL7, and N3C. Our standards for representing biomedical concepts have been widely adopted, including those for human phenotypes (e.g., HPO, GA4GH Phenopackets), diseases (NCIt, Mondo, ICD-11), genes (Gene Ontology), anatomy (Uberon), and molecular variation (GA4GH VRS). We have developed standards and tools to address data provenance (SEPIO), contributions (Contributor Attribution Model), licensing barriers (Data Use Ontology, Reusable Data Project), and connectivity (Linked data Model Language, LinkML). We will build on our previous work, collaborative skills, and technical knowledge to develop a framework to enable the harmonization of standards across biomedical domains. We will form working groups with representatives of the Data Generation Projects (DGPs) to document use cases and synthesize data standard requirements. We will provide protocols and training for specifying standards, and provide concierge services in support of all deliverables and activities. We will create a version-controlled Bridge2AI Standards Registry to inventory standards for use by the DGPs, specified in the modality-agnostic LinkML framework, discoverable through the interactive Standards Hub, and automatically exportable to technical artifacts through our Data Transformation Toolbox. We will build a Standards Evaluation Dashboard for assessment and discovery of standards in datasets from Bridge2AI Data Generation Projects. We will promote best practices in the transparent and responsible sharing of datasets and ML models through DUO, Datasheets, and Model Cards.