Scientific data is produced at an extraordinary and accelerating pace. Genomics, brain imaging, atmospheric modeling, and high-energy physics experiments collectively generate petabytes of data every year. But generating data is not the same as preserving its usefulness. Without sustained, deliberate effort, data silently degrades — not in bytes, but in meaning. Variables lose their definitions. Instruments lose their context. File formats become unreadable. People leave, and their knowledge goes with them.
Data curation is the practice that prevents this. It is the ongoing, active management of data throughout its lifecycle to ensure the data remains accurate, intelligible, accessible, and useful — not just today, but years or decades from now.
What Is Data Curation?
The word curation comes from the Latin curare, to take care of. In a museum, a curator acquires, organizes, contextualizes, and preserves objects so they remain meaningful to future audiences. In science and information management, a data curator does the same for data.
The Digital Curation Centre (DCC), one of the leading international bodies on the subject, defines digital curation as:
“The active management and appraisal of digital information over its entire life cycle.”
More practically, data curation encompasses all of the activities required to keep data meaningful, usable, and trustworthy over time:
- Organizing — arranging data into coherent, navigable structures
- Describing — writing metadata that explains what the data is, where it came from, how it was collected, and how it should be used
- Standardizing — converting data into community-standard formats and vocabularies so it integrates with other datasets
- Validating — checking for errors, inconsistencies, duplicates, and missing values
- Enriching — adding context, cross-references, and links to related datasets or publications
- Preserving — ensuring long-term storage, format migration, and integrity verification
- Enabling access — applying appropriate licenses, access controls, and discovery mechanisms
Data curation is sometimes confused with related terms. Data management is broader — it encompasses strategy, governance, and infrastructure. Data archiving is narrower — it refers to the long-term storage of data that is no longer actively used. Curation sits in between: it is active, ongoing work, not a one-time deposit.
Crucially, data curation is not the same as simply publishing data. Depositing a ZIP file to a repository with no metadata, no readme, and no license is not curation. Curation ensures that what you publish is actually usable by someone else — ideally someone with no prior knowledge of your project.
What Is a Data Curator?
A data curator is a professional who takes responsibility for managing, describing, and maintaining data collections. In research settings, data curators are the people who bridge the gap between scientists who generate data and the infrastructure and communities that need to use it.
The role is genuinely hybrid: it requires both domain knowledge (to understand what the data represents) and technical skill (to work with databases, metadata schemas, APIs, and file systems at scale).
Core Responsibilities
Metadata creation and standards. A data curator writes or enforces metadata that describes the data — its provenance, collection methods, units, controlled vocabularies, and relationships to other datasets. They select appropriate metadata schemas (Dublin Core, DataCite, domain-specific standards like BIDS for brain imaging or MIMARKS for microbiology) and ensure compliance.
Data quality assurance. Data curators review incoming data submissions for errors, inconsistencies, missing values, and format violations. In large repositories, this involves automated validation pipelines as well as human review of edge cases and outliers.
Format standardization. Raw data arrives in many formats — instrument-specific binary files, lab-specific spreadsheets, legacy formats. Curators translate data into open, standardized formats that downstream tools can process without custom conversion code.
Provenance documentation. Curators record the chain of custody for data: who collected it, with what instruments, using what protocols, processed by what software, at what versions. This documentation is what makes a dataset scientifically credible and reproducible.
Access and licensing. Curators apply appropriate licenses (Creative Commons, government use licenses) and configure access controls — determining what data can be fully public, what requires authentication, and what must be restricted under privacy or consent agreements.
Long-term preservation. Curators monitor stored data for format obsolescence, storage degradation, and bit rot. They migrate data to newer formats or storage systems as needed, and verify the integrity of stored files using checksums and fixity checks.
User support. In research repositories, data curators often serve as the human interface between depositors and the archive — helping researchers understand submission requirements, answering questions about access, and troubleshooting technical problems.
Skills and Background
Data curators typically combine:
- Domain knowledge in the scientific area covered by the repository (biology, neuroscience, chemistry, physics)
- Knowledge of metadata standards and ontologies
- Experience with databases, file systems, and scripting or programming
- Understanding of data lifecycle concepts and information science principles
- Communication skills to work with both scientists and infrastructure engineers
The role has historically been associated with library and information science, but modern research data curation increasingly requires hands-on computational skills — the ability to write validation scripts, work with REST APIs, manage cloud or HPC storage, and contribute to ingestion pipelines.
Data Curation at PSC
The Pittsburgh Supercomputing Center (PSC) is home to national-scale scientific data infrastructure where data curation is a core operational function, not an afterthought.
Brain Image Library
The Brain Image Library (BIL) is a national public resource funded by the NIH BRAIN Initiative that enables neuroscientists to deposit, analyze, mine, and share large brain microscopy image datasets. BIL stores petabyte-scale collections of images from mouse, marmoset, human, and other species — data generated by labs across the country and the world.
Data curation is central to BIL’s mission. Researchers who submit data to BIL work with PSC’s data curation team to:
- Convert raw image data to community-standard formats compatible with neuroimaging analysis tools
- Complete structured metadata describing acquisition parameters, specimen preparation, and imaging conditions
- Validate submissions against the BIL schema before ingestion
- Receive persistent identifiers (DOIs) for their datasets
Without curation, a petabyte of images is a petabyte of files with no context. With curation, it becomes a collection that other researchers can discover, understand, and reuse in entirely new studies.
HuBMAP and SenNet
PSC contributes to the Human BioMolecular Atlas Program (HuBMAP) and SenNet, both of which involve distributed data curation across multiple institutional partners. These programs require that data submitted by labs around the country meet community standards for metadata, format, and provenance — work coordinated by dedicated data curation staff across the consortium.
HuBMAP in particular has invested heavily in defining what a “curated dataset” means for human tissue data: controlled vocabularies for cell types and tissue locations (using the ASCT+B tables), standardized pipeline outputs for assay types (CODEX, Visium, bulk RNA-seq, etc.), and metadata fields that enable cross-study analysis. That infrastructure does not exist without curators to design, enforce, and maintain it.
Why Data Curators Matter
Data without curation is data at risk. Studies of scientific data reuse consistently find that the majority of datasets deposited without curation cannot be independently reanalyzed — because the metadata is insufficient, the formats are proprietary, or the provenance is undocumented. A 2018 survey found that only 25% of datasets published alongside journal articles could be successfully accessed and reused by other researchers.
Curation makes FAIR real. The FAIR principles — Findable, Accessible, Interoperable, and Reusable — describe what well-managed data looks like. But FAIR does not happen automatically. Data curators are the people who implement FAIR in practice: they create the metadata that makes data findable, the identifiers that make it accessible, the standards that make it interoperable, and the documentation that makes it reusable.
Scale demands human expertise. As data volumes grow into the petabyte range, automated validation catches many problems — wrong file types, missing required fields, format violations. But automated tools cannot replace domain judgment. Deciding whether a metadata value is biologically plausible, whether an outlier is an instrument error or a real finding, or whether a dataset is sufficiently documented to support independent replication requires a person with both scientific knowledge and curatorial skill.
People leave; curated data stays. One of the most persistent problems in academic research is knowledge loss when people move on. A well-curated dataset is documented thoroughly enough that someone who joins the lab five years from now can understand and build on work done today. Without that documentation, the data may survive on a server somewhere but be functionally inaccessible to anyone who was not part of the original project.
Reproducibility depends on it. The reproducibility crisis in science is partly a data and methods problem: studies cannot be reproduced when the underlying data and code are unavailable, undocumented, or not preserved. Rigorous curation — archiving data with complete provenance, software environments, analysis parameters, and version information — is what makes reproducibility structurally possible rather than aspirational.
Funders now require it. NIH, NSF, and major international funders increasingly mandate data sharing with appropriate metadata and documentation as a condition of funding. The 2023 NIH Data Management and Sharing Policy requires that researchers submit and adhere to data management plans. Meeting those requirements well — not just technically but in a way that produces genuinely reusable data — is a curation problem.
A Growing Profession
Data curation is increasingly recognized as a professional discipline with its own career paths, certifications, and communities of practice. Organizations like the Research Data Alliance (RDA), the Digital Curation Centre (DCC), and the Data Curation Network (DCN) support practitioners and develop community standards.
The Data Curation Network, a collaborative of research institutions and libraries across North America, has developed CURATED — a step-by-step curation workflow for data submitted to institutional repositories: Check, Understand, Request, Augment, Transform, Evaluate, Document. The framework is widely used as a training and quality assurance tool for data curators across disciplines.
As open science mandates expand and data volumes grow, demand for skilled data curators will only increase. It is a role that sits at the intersection of science, information management, and computing — and it is one of the more consequential jobs in the modern research enterprise.
References
-
Digital Curation Centre (DCC). What Is Digital Curation? https://www.dcc.ac.uk/digital-curation/what-digital-curation
-
Wilkinson MD, Dumontier M, Aalbersberg IJ, et al. The FAIR Guiding Principles for scientific data management and stewardship. Scientific Data. 2016;3:160018. https://doi.org/10.1038/sdata.2016.18
-
Brain Image Library. BIL Data Submission Guidelines. https://www.brainimagelibrary.org/submission.html
-
HuBMAP Consortium. Data Portal. https://portal.hubmapconsortium.org
-
Data Curation Network. CURATED Workflow. https://datacurationnetwork.org/outputs/workflows/
-
NIH Office of Data Science Strategy. NIH Data Management and Sharing Policy (effective January 25, 2023). https://sharing.nih.gov/data-management-and-sharing-policy
-
Cousijn H, Feeney P, Lowenberg D, et al. Bringing Together All Research Products: How Adoption of a Minimal Metadata Set Can Enable Open Science. Research Data Alliance. 2018. https://www.rd-alliance.org/
-
Whyte A, Tedds J. Making the Case for Research Data Management. DCC Briefing Papers. 2011. https://www.dcc.ac.uk/resources/briefing-papers
-
Research Data Alliance. https://www.rd-alliance.org
-
SenNet Consortium. Mapping Senescent Cells Across the Human Body. https://sennetconsortium.org