# Governing data with FAIR and stewardship metrics
A pharma sponsor asks a simple question about your multi-omics repository: "How many of our 4,200 datasets can an outside collaborator find, access, and reuse without emailing your team?" If the answer is a shrug, you have a governance problem, not a science problem. In 2026, that gap is what stalls collaborations, delays regulatory submissions, and burns grant renewals.
This lesson turns FAIR from a poster on the wall into KPIs you can report to sponsors and regulators.
What FAIR actually means
FAIR is a set of data principles published in 2016 (Wilkinson et al., in *Scientific Data*). It stands for Findable, Accessible, Interoperable, Reusable. It is about machines and humans being able to work with your data, not about making everything public. FAIR is not the same as open.
Findable: the dataset has a persistent identifier and rich metadata (data about the data: who, what, when, how measured).
Accessible: you can retrieve it through a standard protocol, with clear rules for who gets in.
Interoperable: it uses shared vocabularies and formats so a proteomics file and a genomics file can be joined.
Reusable: it has a clear license, provenance (the record of where data came from and how it was processed), and enough context to trust it.
Multi-omics means combining layers: genomics (DNA), transcriptomics (RNA), proteomics (proteins), metabolomics (small molecules). Each layer has its own file formats, identifiers, and repositories. A single patient sample can generate a VCF file (Variant Call Format, for genetic variants), a FASTQ file (raw sequencing reads), and mass-spectrometry output for proteins. Without shared metadata, these never link back to the same biosample.
Metadata standards you will actually cite
Do not invent your own schemaschemaA schema is the formal blueprint that defines how data is structured, named, typed, and related within a database, file, or message.View full definition →. Regulators and sponsors trust community standards. Name the real ones:
ISA framework (Investigation, Study, Assay): a model for describing experiments.
MIAME / MINSEQE: minimum information standards for microarray and sequencing experiments.
HL7 FHIR (Fast Healthcare Interoperability Resources): the standard for exchanging clinical and health data, relevant when omics links to patient records.
CDISC SDTM (Study Data Tabulation Model): required by the US FDA (Food and Drug Administration) and Japan's PMDA for clinical trial submissions.
Ontologies: EFO (Experimental Factor Ontology), MONDO (disease ontology), and Uberon (anatomy) give you controlled vocabularies so "breast cancer," "BRCA," and "mammary carcinoma" mapmapUsing software to automate repetitive marketing tasks and campaigns, enabling personalisation at scale across channels like email, web, and social.View full definition → to one code.
Public reference repositories that already implement FAIR: the European Bioinformatics Institute's EMBL-EBI resources (including the European Genome-phenome Archive, EGA, for controlled-access human data), and the US NCBI's dbGaP and GEO.
The stewardship metrics
Stewardship is the ongoing job of keeping data governed: assigning owners, enforcing standards, and measuring compliance. You cannot manage what you do not count. Here are the KPIs that matter, grouped by FAIR pillar.
Findability KPIs
PID coverage: percentage of datasets with a persistent identifier (a DOI, Digital Object Identifier, or an accession number). Target: aim for 100 percent for anything shared externally.
Metadata completeness: percentage of required metadata fields populated against your schemaschemaA schema is the formal blueprint that defines how data is structured, named, typed, and related within a database, file, or message.View full definition →.
Search hit rate: share of internal searches that return the intended dataset in the top results.
Accessibility and reuse KPIs
Access request turnaround: median days from a controlled-access request to a decision. For human genomic data governed by a Data Access Committee, this is a real bottleneck.
Reuse count: number of distinct downloads or citations per dataset.
License clarity: percentage of datasets with a machine-readable license.
A worked calculation: the FAIR completeness score
Say your schemaschemaA schema is the formal blueprint that defines how data is structured, named, typed, and related within a database, file, or message.View full definition → requires 10 mandatory metadata fields per dataset. You audit a sample of 200 datasets from a repository of 4,200.
77 percent completeness tells the sponsor exactly where you stand. Now weight it: if the 3 findability-critical fields (identifier, title, assay type) are only 60 percent populated, your findability is worse than the headline suggests. Report the weighted view, not just the average.
Turnaround, a second worked example
Median access-request turnaround is a stewardship metric regulators notice. If in Q1 you logged decision times of 3, 5, 8, 12, and 40 days, the mean (13.6 days) is dragged up by one outlier. The median (8 days) is the honest number. Always report medians for turnaround, and track the 90th percentile separately so slow cases stay visible.
Benchmarks and estimates (as of early 2026)
Hard cross-industry FAIR benchmarks are scarce, so treat all figures as rough estimates, not audited standards.
Studies of published life-science datasets have repeatedly found that a large share lack sufficient metadata for reuse. A widely cited 2020 analysis in *PLOS Biology* on reproducibility, and multiple FAIR maturity assessments since, suggest most legacy datasets fall well below full FAIR compliance. Assume your untended legacy repository is closer to 40 to 60 percent complete than to 90 percent until you measure.
The FAIR maturity indicators from the RDA (Research Data Alliance) give you a structured self-assessment. Tools like FAIR evaluation services help score datasets against community standards.
In the EU, the push is reinforced by policy: Horizon Europe grants require FAIR data management plans, and the European Health Data Space (EHDS) regulation, adopted in 2025, sets rules for secondary use of health data across member states. In the US, the NIH Data Management and Sharing Policy (effective January 2023) requires funded researchers to plan for data sharing.
Set your own baseline, then track improvement quarter over quarter. A realistic first-year target: move findability (PID coverage plus critical-field completeness) from baseline to 90 percent for all externally shared datasets.
🎬 [VIDEO: "The FAIR Guiding Principles for scientific data" - youtube.com - a concise explainer of the four principles with life-science examples]
Knowledge check
1. A team argues that making their multi-omics repository FAIR requires publishing all datasets openly to the public. Why is this reasoning flawed?
2. The sponsor's question, 'How many datasets can an outside collaborator find, access, and reuse without emailing your team?', is framed in the lesson as revealing what kind of problem?
3. Why does multi-omics data pose a particular challenge to the 'Interoperable' principle of FAIR?
MULTIPLE CHOICE
4. Select ALL correct answers. Which elements are essential to satisfying the 'Reusable' principle of FAIR?
Select all the correct answers.
MULTIPLE CHOICE
5. Select ALL correct answers. Turning FAIR from 'a poster on the wall' into governance you can report on requires which of the following?
Select all the correct answers.
Building the governance operating model
Metrics without owners are decoration. Assign roles:
Data stewardData stewardA business-side owner responsible for the quality, consistency and appropriate use of data in their domain.View full definition →: owns a data domain (for example, all proteomics assays), enforces the schemaschemaA schema is the formal blueprint that defines how data is structured, named, typed, and related within a database, file, or message.View full definition →, and answers audit questions.
Data custodian: runs the storage and access infrastructure.
Data Access Committee (DAC): for controlled-access human data, decides who may reuse it under what terms. This is a regulatory expectation, not optional, when handling identifiable genomic data.
Governance in practice: a repository workflow
1. On ingest, reject any dataset missing the mandatory fields. Make the schemaschemaA schema is the formal blueprint that defines how data is structured, named, typed, and related within a database, file, or message.View full definition → a gate, not a suggestion.
2. Auto-assign a PID at ingest so findability is never retrofitted.
3. Map free text to ontologies (EFO, MONDO) so search and interoperability work.
4. Tag the license and access tier (open, registered, controlled).
5. Log every access request and download so reuse and turnaround KPIs compute automatically.
Connecting to regulation
When you submit to the FDA or EMA (European Medicines Agency), provenance and traceability are the whole game. Under 21 CFR Part 11 (the US FDA rule on electronic records and signatures) and EU Annex 11, you must show audit trails: who changed what, when, and why. Your stewardship metrics double as compliance evidence. A complete provenance chain and a machine-readable license are exactly what an inspector or a sponsor's audit team wants to see.
The GDPR (General Data Protection Regulation) angle matters too: FAIR "Accessible" never means bypassing consent. For human data, "as open as possible, as closed as necessary" is the operating principle. Controlled access through a DAC is how FAIR and privacy coexist.