The human exposome is the totality of exposures a person receives over their life course and the impact those exposures have on the individual’s health. In the Summer and Fall of 2022, the National Institute of Environmental Health Sciences (NIEHS) conducted a series of workshops, called the Catalytic Workshop Series,¹ to understand the current state of exposomics research and to guide future research and investments in enabling capabilities. NIEHS issued ambitious goals for this workshop as “we will explore what it means to conduct exposomics experiments, develop new tools, techniques, and technologies, share data for maximizing in silico experimentation, and cultivate the research continuum from fundamental to population health. All this work will contribute to developing a framework for demonstrating the value of the exposome in environmental health.”

During the workshop series, five major areas were identified as necessary to operationalize exposomics. These five areas can be broadly classified as dealing with ‘what to measure’, ‘how to measure’, ‘share and harmonize’, ‘integrate, analyze, and interpret’ and ‘translate and impact.’ Several concepts were considered as especially important by participants for near-term focus, including the goal of developing a data ecosystem in which harmonized data can be found, accessed, and shared through sustained and interoperable data repositories. This sentiment is aligned with recent articles^2-4 which have identified a need for integrated solutions for combining internal and external exposomics research, as well as highlighting an urgent need for an information exchange clearinghouse to facilitate sharing of tools and data.

This paper is an outcome of the workshop series discussions with the aim of developing a framework that enhances data exchange for exposomics research. The envisioned data ecosystem framework would facilitate access to and integration of a vast amount of diverse longitudinal data including general external influences (pollution; weather and social context); external individual-specific factors (diet, infections, self-selected chemical intake); internal individual-specific constituents (metabolic byproducts, microbiome derivatives, inflammatory mediators, stress hormones, etc.) that contribute to the onset and progression of disease; and measures of personal (behaviors, screening, treatment, outcomes); and population health (disease rates, years of person life lost, longevity) and health care (access, costs, quality). The data ecosystem would enable priority concepts brought forward in the workshops, including focuses on understanding the exposome at a community level, advancing exposome-wide association studies (ExWAS) and functional exposomics, and advancing the research and application of nutritional pharmacology and precision nutrition to modify the effects of exposures. More generally, the data ecosystem would facilitate a broad range of exposome research by ensuring that the data sets needed by researchers meet the FAIR principles, which states that the data should be Findable, Accessible, Interoperable, and Reusable.⁵

We consider that such a data ecosystem should be developed with an international scope enabling sharing of data, tools and workflow pipelines across multi-disciplinary sciences, diversity of analytes, and linking studies conducted across continents and merging data from existing and future longitudinal studies. The data ecosystem would entail a federation of data resources that already exist worldwide and across subfields but requires an adoption of common protocols for data access and sharing, adoption of semantic standards to guide the collection, storage, analysis, and leveraging of data and data resources, such as programming languages, packages, algorithms, and cloud-computing services needed to create and maintain it. The effort can be hastened through development of a consensus, international data governance strategy that adapts guidelines for proposing, adapting, implementing, and evaluating processes (feasibility) and outcomes (high quality, rigorous, and reproducible data). Guidelines are needed for an array of tasks, including adoption of common terminologies and ontologies; identification of data sources; steps for data collection, linkages, transfer, storage, and security; risk management; and addressing confidentiality of protected health information including privacy risks related to geo-spatial data. Much like the Human Genome Project, it will require the establishment of a consortium of transdisciplinary researchers, data stewards and data scientists, and technologists from across the world as well as the support, leadership, and cooperation of funding agencies.

We believe that such an ecosystem will not be bound to a single site or even a single continent. The system would be a federation of federations that can interlink data across multiple sites. Such a loosely coupled structure would help in maintaining autonomy but still be guided by FAIR principles to share across repositories. The need for interoperability across such loose coupling will require a strong set of standardizations which we believe is possible given the early stage of the exposomics field. Institutions such as the HHEAR Data Center⁶ and the Exposome Explorer⁷ are promoting such common standards for data depositions. Where standardization is not feasible, developing crosswalks would be possible and helpful as we tackle integrating data across multiple disciplines.

Such an effort is tractable and timely. The existing research data ecosystem is used daily by researchers in environmental health to find, access, and work with a diversity of data. Directed efforts are already improving the ecosystem by providing needed funding to sustain data repositories and knowledge bases, to develop standards and related tools, to create libraries of common data elements, and to create common data sharing protocols and technologies. Efforts, such as the European Human Exposome Network (EHEN), the Global Alliance for Genomics and Health (GA4GH), and the NIH Cloud Platform Interoperability (NCPI) project are already developing and providing federation capabilities. The exposome community has the opportunity to work with and build upon these efforts to ensure the emergence of a data ecosystem that serves its use cases and needs.

Workshop participants recommended the creation of an Exposome Community of Practice (CoP) to promote and foster research and impact along multiple avenues, such as supporting workforce development and advancing new data science and statistical methodologies. This paper presents several recommendations, including the utilization of the CoP alongside other community, coordination, and technical activities that aim to evolve the current data ecosystem to the envisioned ecosystem. To inform these recommendations, the paper first presents a high-level overview of the use cases the ecosystem may serve, data collections to consider as foundational, an overview of data federation and its challenges, current opportunities to build upon, and then proceeds with a listing of recommendations.

Exposomics incorporates data from many scientific domains, each with its own set of characteristics and minimal information standards and needs to align with the FAIR data principles. Table 1 provides a brief description of diverse data domains that form the basis of exposomics research. The data sources, opportunities, and challenges listed in Table 1 can inform and prioritize efforts to develop foundational data sets, tools, and policies for the ecosystem.

In addition to considering the variety of exposomics data types, a FAIR-aligned data ecosystem requires differentiation between the levels of data that are shared. The NIH Genomics Data Sharing Policy²⁵ differentiates five-levels of data based on level of processing and aggregation. Level 0 represents raw instrument data, level 1 represents data after initial transformation from the raw format, level 2 is data that has been cleaned and undergone quality assessment, level 3 data has been processed to identify key features, and level 4 data has been integrated with other related information. A similar rubric needs to be applied to exposomics data. For example, ambient air pollution data obtained directly from an air monitor are Level-0 or 1 while -omics data are often Level-3 as these data are typically normalized and annotated. While data at all levels require annotation of the measurement platforms and study characteristics, high levels data also requires information on data processing in order to meet the FAIR standards. Prioritizing the data domains and data levels for sharing will aid the community in allocating the resources available to maintain data repositories and their capabilities.

While federated systems are not new (a NIST reference architecture already exists²⁶ and multiple guides can be found in popular technical press), prioritizing and tailoring components and capabilities for exposomics research has not been done at the community scale yet. Table 2 provides a listing of important components and capabilities by access type and use case types that are characteristic of a mature federated data ecosystem.

Federation is known to be challenging^31-33 as it requires coordination and agreement between different resource providers along multiple dimensions. Providers must adopt common protocols for communications between software systems; common authentication and authorization mechanisms for controlling access to and use of shared resources (including data use agreements and consent agreements); common standards for data and metadata, as well as common protocols for search, retrieval, and cross-platform analysis.

Ensuring operations and maintenance of necessary federation infrastructure is not easy. Computer and network failures can hamper federated services and software typically needs to be continuously updated for new requirements and to deal with changing security issues. Migrating legacy systems and sustaining the budget needed for continued operations is a challenge especially when time-limited funding mechanisms (e.g., grants) are used and where federation goals go beyond the core mission of the funded resources. For example, work to federate exposome and omic data as part of the HELIX project³⁴ was hampered by mismatches between available server infrastructure and project needs. Cloud resources can help in providing appropriate IT resources as well as in addressing security requirements. Recognizing the design and budget requirement on federation of IT infrastructure will be important for success.

The diversity and breadth of the field of exposomics magnifies these other challenges. Exposomics crosses multiple fields and data types, each having their own standards and tools, which can lead to numerous incompatibility issues and a need for interoperability across standards. Diverse fields such as medicine, environmental sciences, and geosciences do not frequently collaborate, and different funding sources and priorities can hamper such collaborations. The exposomics community is international which brings challenges due to different policies and laws, especially around privacy, as well as practical challenges in coordination.

Federation, however, is not a one-size-fit-all concept and there is flexibility in addressing these challenges. The US Government Data Federation website (https://federation.data.gov/) defines a data federation project as one “in which a common type of data is collected or exchanged across complex, disparate organizational boundaries.” Data can be exchanged across organizational boundaries without resolving issues of authentication, harmonization can be conducted after exchange of data or limited to data where aggregation is prioritized, the use of application programming interfaces (APIs) to automate data exchange can be gradually introduced, and the migration of legacy systems to conform with data sharing protocols can be phased in over time. A data federation that is responsive to both needs and opportunities can be evolved through communities of practices that provide scientific guidance, by data owners and repository directors who coordinate efforts, by working groups that develop policies and promote best practices, and by funding agencies who provide funding for community processes, targeted investments in capabilities development and operations, and for adoption of standards in scientific workflows.

The exposomics field can take advantage of previous and ongoing investments from prior and existing data sharing efforts. For instance, the NIH has recently invested in a common single sign-on/multifactor authentication service, the Researcher Auth Service (RAS). The long-term benefits in enabling cross-site automation and of easing the access burdens that researchers confront daily through the adoption of technology like RAS arguably outweigh the short-term costs of adoption. Work under world-wide organizations promoting data sharing, such as the Research Data Alliance (RDA) and the Global Alliance for Genomics and Health (GA4GH) are developing protocols, toolkits, and reference implementation for needed capabilities like federated discovery, machine readable consent and usage forms, and data usage ontologies which the exposomics community can contribute to and adopt. Open-source data catalog and data sharing platforms (e.g., Molgenis, Gen3, DataVerse, OSF) can greatly lower costs of deploying new catalogs and repositories and foster greater standardization through easy sharing of platform data/metadata models. The recent development of secure data enclaves that support large-scale studies, such as All of Us and the National Covid Consortium (NC3), offer system architectures, design patterns, and best practices for deploying new enclaves and in cases specific technologies such as for privacy preserving record linkage.³⁵ The NIH funded Cloud Interoperability Program (NCIP), which is focused on interoperability between data resources, provides methods, use cases, and solutions for the exposomics community to build upon.

In the areas of common language and standards, traditional medical and clinical standards such as the FHIR, Observational Medical Outcomes Partnership (OMOP) Common Data Model, and the Patient-Centered Outcomes Research Common Data Model (PCOR-CDM) are already expanding to include more social and environmental determinants of health. The Organization for Economic Co-operation and Development (OECD) has developed harmonized templates that exist in many areas of relevance. Efforts like the European Human Exposome Network (EHEN) metadata working group, the Environmental Health Language Collaborative (EHLC), GO-FAIR, PhenX, NIH Common Data Elements portal, and the NIEHS Disaster Research Response Program (DR2) are disseminating and promoting standards and working to address gaps using tools and frameworks such as CEDAR Workbench³⁶ and ISA-TAB.³⁷ These efforts can be readily leveraged to support the needs of the exposomics community.

Groups within the EHEN are developing technologies of interest for an exposomics data federation, including a metadata catalog built on the Molgenis platform³⁸ and the DataSHIELD platform³⁹ which enables analysis across federated data resources without moving the data from the secure environment of the data provider. This last capability provides a pathway towards broader privacy preserving federated analysis as demonstrated in recent work.⁴⁰ Industry interest in and use of privacy preserving federated learning and artificial intelligence (AI) has sparked research and development in this field (see Torkzadehmahani et al.⁴¹ for recent review) that the exposomics field should seek to adopt into practice.

Access to sensitive data is still hampered by a diversity of data and the existing material transfer agreements, data use agreements, and subject consent language around data sharing. There are also needs for developing and adopting templates and the use of forms that are machine readable and harmonizable. While progress in these areas has been slow, the efforts like the International HundredK+ Cohorts Consortium (IHCC) and the GA4GH are seeking to advanced data access procedures that apply across studies at an international level and that should be applicable to exposomics studies. For instance, GA4GH has produced a toolkit for drafting machine-readable consent forms and a data access committee review standards toolkit to promote consistent review criteria and is actively working on improving access under its Regulatory and Ethics Work Stream.

Research and development of natural language models and their application for complex tasks such as text summarization, question answering, and information extraction has undergone tremendous steps forward in recent years, with large language model technologies transitioning to mainstream use. The approaches can be applied to federated systems to improve natural language search, translation of human language, translation of language across scientific subdomains, provide summarization of resources managed across federated systems, as well as to aid in producing informative visualizations such as evidence maps.

The following recommendations can be grouped into two broad categories. The items 1, 2, 3, 6, 7, and 10 focus on community and coordination activities that will require leadership and support from funders worldwide to foster as well as engagement from scientists, data stewards, directors of data repositories, and data scientists/informaticists to align and coordinate efforts. The items 4, 5, 8, and 9 focus on the piloting, development, upgrading, and maintenance of cyberinfrastructure to support the management, sharing, and usage of data sets. Existing funding call, such as the NIH funding for established Data Repositories and Knowledgebases, will help to support these efforts although it is likely that an additional support will be needed. Achieving these ambitious goals also calls for coordinating the funding across nations and individual federal agencies to maximize the return on investment and reduce duplication.

Feedback from the 2023 exposomics workshop indicated a clear community need for a federated exposomics data ecosystem. While non-trivial challenges exist in creating this ecosystem, new data sharing policies and work to build data sharing ecosystems within the life sciences and other scientific communities can provide a foundation for success and advance exposomics research, if there is adequate support to enable an exposomics community wide effort.