The NIH Common Fund has supported multiple programs that have resulted in the creation of numerous data coordination centers (DCCs) that house diverse data and resources. In order to facilitate the ability of researchers to find information across DCCs, the Common Fund Data Ecosystem (CFDE) was formed. The CFDE provides a centralized resource managed by the CFDE Coordinating Center where metadata about DCC data assets is stored. The CFDE Portal enables search of this metadata via web-based faceted queries. The Ontology Working Group within the CFDE has established a process for choosing standards to use for the capture of this metadata from DCCs. Multiple ontologies and controlled vocabularies were chosen and are now in active use by the DCCs to submit metadata to the CFDE centralized resource. As of this writing, there are ~4.5 million file records, 2,700 subject records, and more than 1.7 million biosample records linked to ontology or controlled vocabulary terms in the CFDE resource. ontology, Common Fund, metadata harmonization, metadata integration
The NIH Common Fund was formed in 2006
to provide a mechanism to fund initiatives that do
not fall under the purview of a single NIH institute
or center. Common Fund programs must be
transformative, catalytic, synergistic,
crosscutting, and unique
(https://commonfund.nih.gov/). Past and present
Common Fund programs have resulted in the
creation of numerous data coordination centers
(DCCs) that house the data and resources
produced by a given Common Fund program.
These DCCs generally provide tools for searching
and viewing datasets produced by the program
and often also provide analysis tools and other
resources. In order to facilitate the ability of
researchers to find information relevant to their
research that might be housed in multiple DCCs,
the Common Fund Data Ecosystem (CFDE) was
formed. As of this writing, 11 Common Fund
program DCCs participate in the CFDE. A current
full list is maintained on the CFDE Portal
(https://app.nih-cfde.org/) (
Our overarching goal throughout our efforts to develop a metadata capture system for the CFDE was to take a pragmatic approach to the collection of metadata from each DCC such that we could facilitate cross DCC queries. We did NOT want to attempt to unify all vocabularies or standards in use by any DCC as this is perhaps an impossible task and was certainly not in scope for our project. We also did NOT want to endeavor to capture every piece of metadata information stored at each DCC as this would have been duplicating the function of the individual DCCs, and that was not our mandate. What we wanted to do was find a way that all of the DCCs could contribute metadata that would be useful to a large swath of researchers at a level of granularity sufficient for users to identify datasets of interest. To accomplish this we had to accept that our capture of information would be imperfect and incomplete, keeping in mind that our goal is not to replicate the work of the DCCs, but rather to provide pointers and guideposts for researchers to find the resources provided by the DCCs.
Our process for choosing which ontologies and CVs to use for metadata harmonization as well as for maximizing the utility of those ontologies and CVs for all DCCs included several steps which are described below and in Figure 1 above.
• Choose which metadata types to focus
on. The types of metadata associated with
datasets at each DCC are extensive.
However, as mentioned above, our goal
was not to capture everything available at
each DCC but rather to focus on metadata
elements that would provide the most
utility as search criteria for the maximum
number of researchers. Based on use
cases established for various user
profiles, we chose an initial 12 types of
metadata to capture. These are listed in
Table 1. Over time, we expect to expand
the list of metadata types included in the
CFDE.
• Survey DCCs regarding current use of
ontologies and CVs for those metadata
types. DCC representatives were asked to
fill in an online survey that asked about
their use of ontologies and CVs for the
capture of metadata. Six out of nine DCCs
(that were participating at that time, now
the number of DCCs is 11) responded to
the survey. The survey consisted of
questions asking what data types, assay
types, data formats, etc. were being
captured by each center and what, if any,
ontologies or controlled vocabularies
were being used. DCCs provided answers
as free text.
• Identify candidate ontologies and CVs
for those metadata types. There are
hundreds of ontologies being used in the
biological research community, including
many that cover the same conceptual
areas. We employed the NCBO BioPortal
tools (
questions, o have some level of community buy-in as measured by BioPortal “Acceptance Score” and GitHub issue activity (new issues being submitted recently), o conform to community conventions on ontology and vocabulary development, o provide mappings to other related ontologies/vocabularies, as relevant Discuss and reach consensus. Candidate ontologies were discussed in working group meetings to reach consensus.
Write and circulate a “Request For Comments” (RFC). Once the working group reached a decision, an RFC was written that described the metadata type in question, the ontology/CV that was chosen, and any other information needed by DCCs for correct usage. The RFC was circulated first within the OWG and then throughout the entire CFDE for comment and revision before becoming a final policy. RFCs are versioned and, as needed, revisions to the RFCs will be made.
Facilitate use of the chosen ontologies
and CVs. Some DCCs had not used
ontologies or CVs for storage of metadata
or had been using different ontologies or
CVs than those chosen through the
process above. In addition, the
submission of metadata to the CFDE
central resources in the form of ontology
or CV terms was a process new to the
DCCs. Therefore, OWG members
engaged in helpdesk activities to support
DCCs as they converted (as needed) their
metadata to the OWG standards and
submitted them to the CFDE central
repository. In addition, there were
occasions when DCCs needed terms that
did not yet exist in the chosen ontologies.
In these cases, the OWG facilitated the
•
development of the needed terms by
liaising with the ontology developers,
shepherding new term requests through
the development process, and tracking
term status. In some cases, the chosen
ontology or CV did not provide updated
releases on a schedule rapid enough for
the needs of DCCs. In that case, we
employed the InterLex system to make
provisional terms with in-house ids that
can be used until new terms can be
incorporated into official releases of the
relevant ontologies or CVs (
Build a “slim” for the ontology or CV.
DCCs are always encouraged to use the
most granular terms that are applicable to
their metadata as this provides the most
accurate and specific information.
However, visualization of hundreds or
thousands of terms that have been used
within a dataset can present challenges.
Ontology “slims” can solve this difficulty
by providing a way to see a more
highlevel view of a set of annotations (
The current list of ontologies and CVs that have been adopted for metadata capture in the CFDE are listed in Table 1. According to the nature of the individual metadata types, some ontologies have dozens of terms in use, while others have thousands. Table 1 indicates how many unique terms are being used for each ontology as of July 2022. It also displays which ontologies we have built CFDE-specific slims for. For metadata type sex, a small selection consisting of four SnoMed-CT terms have been adopted (10). They are: ‘Indeterminate’, ‘Female’, ‘Male’, ‘Intersex’. These were chosen to reflect values that DCCs currently have in their project metadata. Moving forward, we realize this field and associated vocabulary are inadequate for all potential needs (e.g. transgender people). Thus, we plan to revise this field and its allowed values in light of United States Core Data for Interoperability (USCDI) (https://www.healthit.gov/isa/united-states-coredata-interoperability-uscdi) and Health Level Seven (HL7) (https://www.hl7.org/) standards as they evolve to meet community needs. To capture race and ethnicity we will be using the Standards for Maintaining, Collecting, and Presenting Federal Data on Race and Ethnicity (https://www.govinfo.gov/content/pkg/FR-199710-30/pdf/97-28653.pdf) and will strive for continued alignment with the USCDI.
The CFDE central metadata repository captures metadata about subjects, biosamples, and files as core entities. Currently, there are ~4.5 million file records with links to OBI assay terms, EDAM format terms, and EDAM data terms. In addition, there are more than 2,700 subject records linked to Disease Ontology terms and more than 1.7 million biosample records linked to Uberon anatomy terms. This is just a sampling of the large quantities of metadata currently contained in the CFDE central repository and accessible through the CFDE Portal (https://app.nih-cfde.org/).
Although the use of standards such as ontologies and CVs is crucial for data harmonization, it is not enough. Even when multiple parties are using the same ontology, there can still be inconsistencies in the use of terms for specific situations. For example, we found that even within a single DCC when looking at output files of the same kind, with the same format, and from the same software tool, different curators sometimes chose different data type terms. The problem becomes even more magnified when looking across DCCs. Therefore, in an effort to increase the consistent use of the standards chosen for the CFDE central metadata repository, we are working to identify inconsistent use of terms and then to build term-use guidelines that will address those inconsistencies across DCCs.
In the area of ontology/CV slim development, we will continue to revise the slims we have built for each ontology/CV as more metadata records are submitted to the central repository and as new DCCs join the CFDE so as to maintain the most useful set of general categories for each ontology/CV as possible. We will also build a slim for PubChem since, with more than 73,000 terms in use, a slim will be very advantageous for summary views and comparisons. We will build slims for additional ontologies used in CFDE as needed.
As work on the project continues, we hope to add additional metadata types to our central repository. We will use the above process to identify community standards to use for those datatypes as well.
We wish to acknowledge the NIH for funding much of this work through award OT3-OD02549. We also want to acknowledge all of the members of the Ontology Working Group: David Chen, Keyang Yu, Matt Roth - Baylor College of Medicine; Jared Nedzel - Broad Institute; Allison Heath, Deanne M Taylor, Eric Wenger, Taha M. Ahooyi - Children's Hospital of Philadelphia; Mano Ram Maurya - Department of Bioengineering, University of California, San Diego; Avi Ma'ayan, Eryk Kropiwnicki, John Erol Evangelista, Sherry L. Jenkins, Zhuorui Xie - Department of Pharmacological Sciences, Mount Sinai Center for Bioinformatics, Icahn School of Medicine at Mount Sinai; Daniel Lyman - George Washington University; Andrew Schroeder, Sarah Reiff - Harvard Medical School; Ellen Quardokus, Katy Borner - Indiana University; Robert Carter - Institute for Genome Sciences, University of Maryland School of Medicine; Asiyah Lin, Chris Kinsinger, Erika Kim, George Papanicolaou, Haluk Resat, Olga Vovk - National Institutes of Health; Brian Walsh - Oregon Health and Science University; Diane Eshelman, Phil Blood - Pittsburg Supercomputing Center; Christophe Lambert, Cristian Bologa, Jessica Binder, Vincent Metzger - School of Medicine, Department of Internal Medicine, Translational Informatics Division, University of New Mexico; Soha Hassoun - Tufts University, Medford; Bernard de Bono - University of Auckland, New Zealand; Amanda Charbonneau, Jeremy Walter, Saranya Canchi - University of California, Davis; Srinivasan Ramachandran University of California, San Diego; Jie Liu, Yuanhao Huang - University of Michigan; Steve Mathias - University of New Mexico; SusannaAssunta Sansone - University of Oxford.