Findable, Accessible, Interoperable, and Reusable Data for Maximum Research Value
The FAIR principles, published in 2016 in Scientific Data, provide guidelines for making research data Findable, Accessible, Interoperable, and Reusable. Rather than a standard or specification, FAIR is a set of guiding principles that emphasize machine-actionability—the capacity of computational systems to find, access, and process data with minimal human intervention.
FAIR emerged from the recognition that research data's long-term value depends on more than just availability. Data must be discoverable, obtainable under clear conditions, combinable with other data, and structured for reuse. These principles apply to data itself, to metadata, and to the infrastructure enabling discovery and access.
The first step toward data reuse is discovery. Data and metadata must be easy for both humans and machines to find:
| Requirement | Implementation | Examples |
|---|---|---|
| Persistent Identifiers | Assign DOIs or similar PIDs | DataCite DOI, ORCID for researchers |
| Rich Metadata | Use community standards | Dublin Core, DataCite Metadata Schema |
| Indexing | Register in searchable catalogs | re3data, Google Dataset Search |
| Keywords | Use controlled vocabularies | MeSH, SNOMED CT, ontologies |
DOIs provide globally unique, persistent identifiers for datasets. Unlike URLs that may break, DOIs are maintained by registries and continue to resolve even when data moves. DataCite manages DOIs for research data, providing metadata schemas and resolution infrastructure. Each dataset should have its own DOI, separate from any associated publication DOI.
Once data is found, users need to know how to access it. Accessible means data can be retrieved, not necessarily that it must be open to everyone:
FAIR does not require all data to be open. Sensitive medical data may require controlled access with data use agreements. The key is that access conditions are explicit and machine-readable. "As open as possible, as closed as necessary" is the guiding principle. Metadata describing the data should always be open, even when data access is restricted.
Data often needs to be integrated with other data or used with diverse applications and workflows:
| Domain | Standards | Purpose |
|---|---|---|
| Clinical | HL7 FHIR, OMOP CDM, CDISC | Clinical data exchange and analysis |
| Genomics | GA4GH standards, VCF, FASTQ | Genomic data formats and exchange |
| Imaging | DICOM, NIfTI, OME-TIFF | Medical and microscopy imaging |
| Ontologies | OWL, SNOMED CT, MeSH, HPO | Semantic annotation and linking |
Ontologies provide formal, shared conceptualizations of domains. In medical research, ontologies like SNOMED CT (clinical terms), HPO (human phenotypes), and Gene Ontology enable precise annotation that computers can reason over. Using ontology terms rather than free text dramatically improves interoperability and enables discovery across datasets.
The ultimate goal of FAIR is data reuse. Data should be well-described so it can be replicated or combined with new research:
Several tools and frameworks exist to assess FAIR compliance:
| Tool | Approach | Use Case |
|---|---|---|
| FAIR Evaluator | Automated checks via API | Technical compliance testing |
| FAIRshake | Community assessments | Repository evaluation |
| FAIR Maturity Indicators | RDA working group standards | Formal assessment framework |
| FAIR Self-Assessment | Questionnaire-based | Planning and gap analysis |
The European Commission estimated that FAIR data practices could save €10.2 billion annually by reducing time spent searching for data, avoiding duplicate research, and enabling secondary analysis. Return on investment studies show that for every €1 invested in research data infrastructure, €4 in economic benefits are generated.