Zebrafish metadata: connect records before reuse

Finding a results file six months after a zebrafish experiment does not necessarily reveal what was measured. The reader also needs the animals, tanks, conditions and analysis version associated with it. Metadata describe that context, but they are useful only while their connection to the result remains intact.

Wilkinson and colleagues published the FAIR principles in 2016: research objects should be findable, accessible, interoperable and reusable. The principles concern data as well as tools and analytical workflows. In work using Danio rerio, the practical task is to establish those relationships during the experiment, rather than attempt to reconstruct them when a dataset is shared.

Establish identifiers before files multiply

A tank identifier cannot identify an experimental group indefinitely because its occupants may change. The register distinguishes the cohort, the tank occupied on a particular date, the sample and the acquisition file. Relationships between those objects need to survive transfers and changes in file location.

FAIR principles call for persistent identifiers, rich metadata and qualified links between objects. Within a facility, a coherent register of local identifiers is a realistic starting point. A published dataset can then receive the persistent identifier assigned by the chosen repository. Its local identifiers remain necessary for understanding the individual records it contains.

A mapping table should let someone trace a result back to its original acquisition and context. Testing that route on a few records often reveals conflicting names, undocumented changes or files that cannot be assigned confidently. FAIR does not prescribe a particular software product, so the facility can choose an implementation suited to its resources.

Record biological identity precisely

A line name does not describe the genotype of every animal. Records should identify the relevant alleles, zygosity and confirmation method. ZFIN’s author guidance asks for precise designations and explains that an unspecified allele limits the usefulness of a record.

Phenotype describes an observation, rather than a result inferred from a genetic name. Retaining the measured endpoint, stage and method prevents biological identity from being confused with an experimental finding. If only a subset was genotyped, the record should not suggest that every fish was verified individually.

The 2025 Zebrafishology guidelines add details specific to zebrafish experiments. Origin, developmental stage, maintenance conditions and experimental units need to remain available. A spawning event creates relationships between animals that must not disappear when the resulting files are organised by date or operator. Those relationships are part of the context a later analyst needs.

Connect water records to the observation period

A temperature set point and a measurement taken in a tank are different records. State which is retained, where it was obtained, the unit and its date. An incident between acquisitions should be a dated event linked to the affected cohort, rather than an unconnected note.

Lighting records can include the programme and the conditions relevant to the experiment. A general facility setting is not a substitute for a change in the acquisition environment. Feeding, density, handling and treatments should be documented according to the scientific question; recording metadata does not require making every conceivable detail mandatory.

A practical structure includes a data dictionary defining each field, its unit, permitted values and the meaning of a blank entry. “Not measured”, “unknown” and “not applicable” must remain distinguishable. Otherwise a future user may interpret absence of information as a biological result or combine records that should have been kept separate.

Preserve the route from signal to result

Provenance describes the steps leading from acquisition to the reported finding. Retaining the original file, analytical settings, software version and exclusion rules makes that route inspectable. A corrected copy should remain distinct from the original, with the correction and its reason recorded.

Sampling design also determines what a row represents. Ten observations from one tank do not become ten independent tanks. Relationships between repeated measures, fish and experimental units must survive export. Without them, a later user may repeat the arithmetic correctly while answering a different scientific question.

Agree responsibilities before the project ends. Identify who maintains the dictionary, validates exports, preserves originals and handles an error discovered after deposit. This is a practical organisational recommendation, not a FAIR certificate or evidence that a biological conclusion is correct.

Version information should be specific enough to distinguish the analysis actually performed from the current default configuration. If a threshold or an exclusion decision changes, the team should be able to identify which output used it. Replacing an earlier result without preserving that relationship makes the record less interpretable, even if the replacement is an improvement.

Controlled access can still support reuse

FAIR does not require every file to be freely downloadable. The principles allow authentication where necessary, call for explicit licensing and retain access to metadata when data are no longer available. Describe access conditions clearly enough for a prospective user to know how to request the dataset.

A facility can test its dossier by asking a colleague outside the project to trace one result to biological identity, conditions and processing steps. Use the gaps found in that exercise to decide which records to repair. Also test backup restoration: a well-described link to a lost file cannot support reanalysis.

Keep this exercise separate from scientific validation. A richly documented dataset may still be biased, incomplete or based on an inadequate design. Interoperability makes comparisons easier to organise; it does not make them biologically valid without examining the methods and populations involved.

Conclusion

Zebrafish metadata support reuse when a result can be traced back to animals, conditions and successive transformations. Establishing those connections during the experiment reduces uncertain reconstruction later. A later user should be able to follow that route and understand the limitations and access conditions.

Vetofish support

Vetofish can help define husbandry and health-monitoring fields, review their consistency and organise responsibilities through our advisory service. Our work with research facilities supports this traceability alongside experimental-design review and the institution’s own data-governance arrangements.