Sharing SRP data to reduce environmentally associated disease and promote transdisciplinary research

Programmatic efforts to facilitate data sharing dialogue

To address the increased need to share data in response to strategic plans across many agencies, the SRP and its grantees have been proactive in recognizing the importance of leveraging, integrating, and reusing data. To encourage data sharing, integration, and reuse between SRP grantees, the SRP has made use of time at annual meetings, convening sessions and keynote speakers, to highlight successful data sharing, facilitate conversations, and receive feedback from its grantees on how research findings from one field might be used in other analyses. SRP researchers have also been involved in workshops to promote data integration, including the National Academy of Sciences, Engineering, and Medicine workshop on informing environmental health decisions through data integration (21).

In 2015, the SRP hosted a webinar series, Integrating Data from Multidisciplinary Research, to explore the challenges and opportunities in integrating datasets to solve complex environmental health problems (22). The series, which included presenters from the NSF, the USGS, and the NIEHS Children’s Health Exposure Analysis Resource (CHEAR), introduced the importance of data sharing and highlighted projects that enable data discovery and facilitate sharing of research and expertise. The webinar series provided a way for SRP staff and grantees to see how other EHS researchers are tackling data sharing. The webinars also provided the opportunity to interact with data sharing experts in the field, ask questions, and share experiences. Below we describe strategies that could be employed across the broader environmental health fields using specific examples from the SRP. These data sharing goals are not unique to the SRP and may be shared with the wider research community.

SRP program staff also encourage collaborations between researchers with shared interests through the K.C. Donnelly Externship. The externship provides a vehicle for SRP trainees to collaborate with researchers outside their home institutions while broadening their expertise and skills. The opportunity also facilitates research collaborations and data sharing among grantees (23).

Developing an SRP dataset library

Datasets are important products of scientific research. Since the inception of the program in 1987, SRP grantees have produced more than 12,000 publications as part of their SRP-funded research projects, but only a small percentage of these publications have associated datasets available in public repositories. It is likely that much of the data generated as part of the research described in these publications are so-called dark data, where the data are either not shared at all or lost in supplementary information (24). According to an International Data Corporation study, more than 90% of digital data are dark data (25).

Datasets archived in journal websites as supporting files are not indexed, and therefore are effectively hidden from discovery. Another study found that only about 20% of data availability statements in publications indicate that data are deposited in a repository (26).

Supplemental data that may no longer have value in one field may be useful to another research project and should therefore, in many cases, also be publicly available, meaning that they are available online for anyone to use. There is likely to be additional data beyond what is captured in underlying primary publications, including study data reflecting null findings that are often not published, which could be relevant and repurposed by another researcher to address a different research question. To begin to make data more findable and accessible, and to provide the research context from which the data were generated, the SRP created and maintains a searchable catalog of publicly available datasets from SRP-supported publications (27).

The dataset listing is housed on the SRP website, which is generally used to communicate major research findings and events from across the program. The datasets are linked to their associated SRP projects and users can browse or search dataset tables by dataset title, data type, data repository, organism, dataset accession number, and SRP project(s), making the data more findable and accessible. Each dataset record contains metadata, including data type, experiment type, organism, and a link to the actual dataset and information (e.g. abstract and project updates) on related SRP publications and projects. This added context about how the data were generated within an SRP center provides users with additional information that is not normally available in a dataset repository. Users can also search the SRP publications associated with the datasets for specific terms, such as a target chemical, to find datasets associated with research-specific topics. This is not a data repository, but a listing of SRP publication-associated datasets, which encourages access to shared SRP data and provides one way to increase findability of SRP data. Specifically, these datasets are connected to a description of the research project from which the data were generated, providing critical additional information on the context of the dataset that may not be clear in a large data repository of diverse datasets.

To populate this dataset listing, PubMed and Clarivate Analytics’ Data Citation Index on Web of Science are regularly searched to identify publications and public datasets associated with SRP funding. In addition, dataset information is provided by SRP grantees to program staff. Datasets are added to the existing SRP database and displayed on the SRP website (3415 datasets included as of January 21, 2020). Examples of data types include DNA and protein sequences, gene expression, genotype and phenotype, proteomics, exposure, environmental science, and clinical trial data. Table 1 provides a summary of the number of datasets by general data type in the SRP database. The database is currently populated primarily by nucleotide sequence and gene expression data. Because SRP research being conducted across the program is much broader in scope, as shown in Figure 1, there is potential to broaden the publicly available dataset data to include much more diverse data types.

Providing researchers and the public a way to find and access SRP-funded data increases the impact and visibility of SRP research and promotes data sharing and collaboration. However, identifying SRP-funded datasets and tracking data sharing can be difficult. While many scientific journals and funding agencies mandate making data freely available to the public, data sharing and the practice of data citation are still in early development. Data mentions and citations vary greatly across research fields and formal citation of datasets using digital object identifiers (DOIs) and data citations is limited, with the majority of datasets archived in the journal’s supplementary materials (28). This is a problem because when data are provided in the supplemental information of a publication, they can be difficult to search, find, and reuse. Datasets that are archived in data repositories and assigned DOIs or accession numbers are much more likely to be cited and therefore tracked.

Enabling data integration through data management cores

In its most recent multiproject center request for applications, SRP added a requirement for a Data Management and Analysis Core (DMAC) (RFA-ES-18-002, https://grants.nih.gov/grants/guide/rfa-files/rfa-es-18-002.html). The primary purpose of the DMAC is to support the management and integration of data across the center, which includes establishing, coordinating, and monitoring steps for collecting, processing, and analyzing data. As it is critical for data scientists to work closely with the subject matter/research domain experts, the SRP intentionally designed the DMACs so this interaction is central to this research support core. For this reason, the DMAC will work closely with project and core leaders to ensure high data quality throughout the data life cycle. The DMAC will also work with project and core leaders to identify opportunities for integrating project and core-generated data with other existing datasets. DMAC leads will also be interacting across the program so that challenges and opportunities can be discussed along the way. The goal of the DMAC is to create shared data standards and systems between biomedical and environmental science and engineering projects, improving the extent to which different projects can exchange and interpret data. By facilitating data integration, the DMAC aims to accelerate the impact of the center’s research (29). Including data management plans that consider data collection for a given research project in the context of the larger center research questions can help to maximize the potential of future data integration among research projects.

Research infrastructure facilitates data sharing and integration

SRP has research infrastructure in place that can facilitate data sharing. Over the years, large cohort studies funded by the SRP have amassed a variety of biospecimens and data on exposures and health effects. These large, ongoing cohorts are a rich source of data that may be leveraged when combined with other studies. SRP researchers also have a wide variety of mechanistic data based on animal studies. Moving forward, SRP researchers have the opportunity to combine epidemiology and exposure data to improve the assessment of risk from environmentally relevant exposures. For example, SRP grantees and colleagues are developing software focused on integrating multiple sources of evidence, including epidemiology and mechanistic data, toward a better understanding of health outcomes (37). SRP researchers are also developing tools to standardize dose-response data modeling and creating computational approaches to predict toxicity values for use in health assessments on a large number of chemicals (38), (39).

Spatial data collected over time on the fate and transport of contaminants may also help to inform what we know about the potential for exposure and health effects. Ongoing studies combine exposure data with health outcome information so we can start to visualize potential links between exposure and disease (40), (41), (42). Other studies are using spatial modeling to characterize the fate and transport of contaminants into the environment (43), (44). It is important to harmonize these different types of spatial data to provide researchers with a comparable view of data from different studies. For example, an NIEHS-funded study linked birth record data to the CalEnviroScreen dataset, which contains spatial data on exposures and environmental effects, to evaluate cumulative exposures related to preterm birth using environmental and social information (45). In another study, NIEHS-funded researchers are working to operationalize built environment measures at varying spatial scales so that environmental data can be linked more appropriately to health outcome data (46).

SRP research infrastructure has also provided an avenue for researchers to move quickly to respond to disasters. In the aftermath of Hurricane Katrina, SRP provided emergency supplemental funding to study health effects. As part of this funding, SRP grantees developed the NIEHS Environmental Health Sciences Data Resource Portal that integrated advances in geographic information systems and data visualization technologies to serve as a national resource to track environmental hazards after disasters. The portal provided a way for NIEHS researchers and partners to work together more effectively and obtain shared datasets and applications. It also functioned as a way to test new technologies to help advance environmental health research (47). Using the portal, SRP investigators provided Gulf Coast leaders with tools and data to monitor health impacts, assess and reduce human exposures to contaminants, and develop remediation strategies (48).

Parallels between health and remediation research

The diversity of SRP research may create opportunities that link findings across a variety of fields, such as health and remediation research. For example, both fields include research related to microbial communities. As with humans, the microbiomes of plants, animals, and the environment play a role in growth, development, and vulnerability to contaminants. Microbiomes are influenced by different environments and this data may inform microbial research in other fields.

In order to integrate microbiome data across fields, there is a need for standardized methods and metadata collection to make robust interstudy comparisons about microbial communities. Efforts are underway to develop policies and priorities for organizing microbiome studies and creating a catalog of microbiomes (49). NIH also created the NIH Interagency Strategic Plan for Microbiome Research FY 2018–2022 to support collaborative research, develop robust and consistent standards, support open and transparent data, and further develop analytical technologies (50).

Many of these efforts focus primarily on the animal and human microbiomes. Given the inclusion of both biomedical and environmental science research within a center, SRP data provides the unique opportunity to take this one step further, with data focused on microbe-microbe interactions in the environment (51) as well as biogeochemical interactions that help specific bacteria degrade pollutants (52). Exploring how microbes break down contaminants in the environment is important to the field of environmental remediation, but could also translate to other fields of research, such as how microbes respond to contaminants in the body. SRP grantees have data on specific contaminants and specific microbes, and researchers may be able to use that data to accelerate research on the human gut microbiome. The gut microbial ecosystem may also depend on where we live (53). Integrating spatial and temporal data with information about the microbes may further our understanding of the human gut microbiome, an area of science that is not well understood.

Privacy

Many environmental health researchers have expressed concerns about privacy of research participants when, for example, cohort datasets are shared (54). It is possible for data to be de-identified before sharing or storing in a public database. However, for many datasets, its usefulness hinges on knowing potentially identifying characteristics. Examples include location data in a pollution study or specific race in studies aiming to identify health disparities. When several attributes are associated with individuals, even de-identified data can be re-identified. A recent study found that 99.98% of Americans could be identified in an anonymized dataset that includes 15 demographic attributes (55). Concerns about privacy are complex and real, and such risks may discourage people from participating in research studies in the future.

Various approaches have been identified to address these concerns, which include adopting privacy-enhancing technologies for data analysis, ensuring data security and improving procedures for controlled access, and making privacy requirements more consistent (56). For example, the NIH All of Us Research Program, an effort to accelerate research and improve health by gathering data from 1 million or more people living in the US, commits to ensuring confidentiality and integrity of all specimens and data. As part of the Precision Medicine Initiative, they are building security practices into development and identifying strong safeguards to ensure privacy. They also reevaluate their methods regularly to keep up with advancing technology (57). Improving procedures that automatically reduce potential identifiers would be critical to make the process of data sharing more efficient and attractive to the research community.

Training

Others have pointed out the large investment needed to train the next generation of scientists to more routinely work with large datasets (58). As biomedical research is accelerating its reliance on computational, mathematical, and statistical methods, supporting training requires new emphases on analytical skills (59). The overarching needs are twofold: bringing more trainees from computational biology into EHS fields to perform data analysis and mining, as well as providing more scientists in the EHSs the knowledge they need to work with both data and data science experts.

For example, trainees with computational skills could develop scripts to automate processes, which would otherwise be executed by laborious and time-consuming step-by-step approaches, making data analysis more efficient. It is also important to address the disparity between scientists without data science expertise and those who have been recently trained with these much-needed analytical skills. Training in data science is not just important for early-career scientists but should extend to mid- and late-career scientists who have a deep expertise in a given area of science and could benefit from a better understanding of data science and how it can increase the robustness of their data. Scientists at all levels from all disciplines must know enough about databases, data storage, and data mining to ask the right questions of their computational biologist colleagues, inform data scientists on whether an approach is good for the field, and work with them to evaluate the data, including recognizing limitations (60). While this training may not provide the skillset needed to independently take on data science approaches, it can serve as a foundation for the subject matter expert to more efficiently partner with the data scientist. In addition to training subject matter experts to understand basic data science, data science experts in key disciplines related to data analysis and integration are also important to ensure data are used thoughtfully and effectively.

To gain a better understanding of how existing data science and environmental health resources (e.g. trainee pipelines, mentors, research) can be used, to identify what is needed, and to build capacity in data science for environmental health researchers, the NIEHS convened a workshop in 2018. The workshop, Developing a Data Science Competent EHS Workforce, brought together data experts from relevant research disciplines to examine data science resources and provide recommendations on how these resources can address EHS-specific training goals in data science training (61).

Ontologies/semantic reasoning

Enabling data sharing, integration, and searching requires semantic standards, such as ontologies, a need that has been recognized by many scientists and scientific organizations (62). The use of ontologies can facilitate data integration by standardizing vocabulary for describing different entities and relationships between them. Ontologies can improve metadata, automatic data verification, and support for queries of data (63). In order to improve data sharing across multiple laboratories, researchers will need to work together to standardize terminology and improve existing ontologies to keep up with the evolution of research. Representation of information using standardized ontologies not only improves data aggregation and integration, but also opens up the use of automatic processing to derive facts that are not expressed in the ontology specifically, known as semantic reasoning (62), (64). Utilization of standardized ontologies also facilitates the development and application of machine learning methods that can integrate data from different datasets to generate new knowledge. Due to the transdisciplinary nature of EHS research, developing a common language may be especially challenging, but it will yield valuable knowledge for many fields beyond EHSs. The effort that will be required to develop ontologies and common languages for these data has benefits reaching a variety of fields.

These benefits are exemplified in the existing Comparative Toxicogenomics Database (CTD), which gathers and analyzes data about chemical-gene interactions and chemical-disease and gene-disease relationships, to construct networks of chemical-gene-disease relationships. The CTD integrates data from real-world exposure studies into the database and includes a standardized vocabulary for diverse study designs and types of data. It also provides much-needed real-world context to exposure information that will enable its use in developing hypotheses about connections between real-world exposures and diseases (65).