Zebrafish studies: randomise before recording

Automated video tracking can produce precise measurements from a poorly allocated experiment. For zebrafish studies, randomisation and blinding therefore belong in the design before recording starts. This guide applies ARRIVE and NC3Rs principles to aquatic research: identify what receives the experimental condition, plan allocation and retain masking where it can protect decisions. Automation is useful, but it does not make those design choices on behalf of the research team.

Identify the independent intervention first

For zebrafish, Danio rerio, the number of animals observed is not necessarily the number of independent experimental units. NC3Rs defines an experimental unit through independent assignment to an intervention. That decision determines what the sample size represents and must precede the choice of statistical analysis.

Consider an experiment comparing tank-level environmental arrangements. Fish within a tank share the condition assigned to that tank. Recording each fish separately does not create additional independent allocations. Similarly, several video clips from one animal provide repeated observations rather than automatically providing several independent experimental units. The analysis must reflect that structure instead of treating every spreadsheet row as another replicate.

In another design, the individual fish may be the unit because conditions can be independently assigned at that level. There is no single answer for every aquatic experiment. We recommend drawing the relationships between animal origin, housing, assigned conditions and measurements before the study begins. If independence remains uncertain, resolve it with statistical input before animals are used.

This diagram can also reveal where practical arrangements differ from the intended design. A treatment assigned on paper to separate tanks may still depend on shared equipment or a common session. Those dependencies need to be examined in context, rather than assumed away because each tank has a separate identifier.

Make allocation reproducible

ARRIVE asks researchers to state whether allocation was randomised and explain how the sequence was generated. An operator’s informal choice of an animal “at random” is not necessarily an adequate allocation method. Randomisation helps avoid systematic group differences, but small samples can still be unbalanced by chance.

For aquatic facilities, a practical record can retain the eligible units, allocation method, resulting sequence and date of generation. The assigned identifier should remain linked to the raw data and final analysis. Renaming a tank or exporting a video file should not break that connection. This is a straightforward way to preserve the design through routine data handling.

Potential confounding also extends beyond initial group assignment. If one condition is always measured in the morning and the other in the afternoon, time becomes inseparable from the comparison. Recording order and location therefore need consideration alongside allocation. Lighting and handling-related stress can also vary with session arrangements. A camera may apply the same tracking algorithm to both groups while the experimental schedule still introduces a systematic difference.

Use blocking for a defined reason

ARRIVE describes randomisation within blocks as a way to organise conditions across relevant subsets of an experiment. Acquisition day, batch or position may be useful factors, depending on the study. Factors used in blocking should also be represented in the analysis. Adding many blocks without a clear rationale does not automatically strengthen a design.

As a practical example, avoid placing every unit from one condition in a single recording session when the study can accommodate conditions across sessions. The allocation within sessions should be planned rather than improvised as equipment becomes available. This is an illustration of the principle, not a universal schedule: the scientific question, facility capacity and dependencies between animals determine the appropriate arrangement.

Age, sex and genotype may also matter. Their role should be considered before results are inspected, rather than selected later because one comparison looks favourable. If the study is intentionally limited to one sex or age range, that boundary needs to remain explicit. Balance between groups is useful but does not replace a description of the animals actually studied.

Blinding has several stages

Blinding limits knowledge of treatment identity where that knowledge could influence decisions. ARRIVE distinguishes allocation, experimental conduct, outcome assessment and analysis. Automatic recording does not demonstrate that every stage is blinded. Staff may still select usable footage, define tracking regions, adjust settings or decide how to handle unusual observations.

Neutral identifiers can help separate those decisions from expectations about a condition. We suggest keeping the code key separately from the files used for assessment. An analyst may need to know which samples form a group without knowing which experimental condition that group represents. The arrangement must be feasible and its limits documented.

Animal care and safety take priority over masking. Some visible characteristics may also reveal a condition despite coding. In such cases, state which stages could not be blinded, why, and what protections remained possible. A precise account is more useful than describing the entire experiment as blinded when the actual workflow does not support that claim.

Agree exclusion rules before looking at the answer

Inclusion and exclusion criteria should be defined in advance, and omitted units or observations should be reported with reasons. A failed video recording, lost identification and a welfare-related withdrawal are different events. Keeping them distinct allows readers and colleagues to understand what happened and how the final dataset was formed.

Analysis decisions should also be documented before group identities are revealed where possible. Software version, tracking settings and manual interventions are part of the evidence trail. A change made after inspecting results is not necessarily invalid, but it should be recognisable as such. Preserve the original files and a record of subsequent changes so the analysis can be repeated.

We recommend testing this record-keeping workflow on a small technical example before starting the study. The purpose is to confirm that identifiers survive acquisition, export and analysis, and that the code key remains controlled. Such a check addresses the data pathway; it does not replace justification of animal numbers or scientific validation of the endpoint.

Precision depends on design as well as measurement

A strong automated study starts with an explicit experimental unit, a documented allocation strategy and protection against avoidable decision bias. These measures do not guarantee a positive finding. They make both positive and negative findings easier to interpret, assess and reproduce.

How Vetofish can help

Vetofish can help align the study plan with aquatic housing, care and animal traceability, working alongside scientific leads and statistical specialists. Early review can identify dependencies and practical decision points, while a clear record of deviations supports interpretation without compromising animal welfare.