Neuronautix · 2026-06-19
When poor data stewardship wastes animal lives. If an animal contributes to science, its data should not die in a drawer.
Damien Huzard, PhD · Neuronautix
The moral architecture
The question must matter. The suffering must be minimized. The number of animals must be justified. The knowledge must be robust enough to justify the cost. This is the moral architecture behind the 3Rs — Replacement, Reduction and Refinement (Russell & Burch, 1959).
A fourth obligation
A mouse can be handled carefully, housed properly, monitored ethically, and approved by every committee — and still be ethically wasted if the data it generates are incomplete, poorly annotated, impossible to reuse, or lost after a single publication.
The unpaid obligation created when animals contribute to research but the research system fails to preserve the full value of that contribution. It is not fraud. It is the result of normal scientific habits that are no longer defensible.
Volume is not value
Home-cage monitoring, video tracking, telemetry, high-resolution phenotyping, omics and imaging now generate vast volumes of data. But data without metadata are not evidence — they are biological residue.
A timestamp without experimental context
A behavioural label without an ethogram
A locomotor trace without strain, sex, age, housing
A raw video without provenance
Device configuration and firmware
Light cycle, protocol version, exclusions
Controlled vocabularies and ontologies
Analysis pipeline and provenance
Reduction, reinterpreted
Reduction is often read too narrowly — the smallest sample size for one study. But it is also about making each animal contribute maximally to cumulative knowledge.
A study that uses the “correct” number of animals but produces unusable data has not respected Reduction. It reduces animals locally while increasing animal use globally — because future researchers must repeat what could have been reused.
Publication bias
A real but unwelcome result is generated
It stays in a drawer, inaccessible to the field
Another lab unknowingly runs the same experiment
The same absence of effect is rediscovered
ter Riet et al. (2012) documented the magnitude and drivers of publication bias in laboratory animal research. In animal research, the cost is paid in lives.
The reproducibility crisis
Cost of irreproducibility (US)
~$28B
Per year, estimated by Freedman et al. (2015) for irreproducible preclinical research in the United States.
Landmark cancer findings
Many failed
Begley & Ellis (2012): industry scientists could not reproduce many landmark preclinical cancer results.
The ethical problem
Non-info.
Not failure — failure is honest science. The problem is preventable non-informativeness: lives spent with no reusable evidence.
The moral question is not “did the experiment work?” — it is “did it leave reliable, reusable evidence proportional to the animal cost?”
ARRIVE · PREPARE · FAIR
Planning and quality control before the study begins. Essential — but a plan is not a reusable dataset.
Defines what to report. A paper can be ARRIVE-compliant and still leave a dataset that cannot be integrated into future work.
Findable, Accessible, Interoperable, Reusable — not an end-of-project label, but a condition of ethical data generation.
FAIR must begin before the experiment. Retrofitting metadata after the fact is fragile, incomplete, and sometimes impossible — context decays once people, devices and software change.
Born-FAIR
The 2024 Minimal Metadata Set (MNMS; Moresis et al.) made this operational — a minimal, ARRIVE 2.0-aligned set to repurpose nonclinical in vivo data. Realistic enough to be filled, rich enough to enable reuse.
Behaviour is exquisitely context-dependent — strain, sex, circadian phase, housing density, enrichment, handling, device layout, group composition, analysis definitions. “Social interaction” in one system may not map onto the same label in another. Without shared metadata, apparent comparability is an illusion. WellFAIR: structure data at the point of capture, not after the project is over.
The invisible evidence base
Pilot data and baseline measures
Control-group observations
Negative and neutral findings
Welfare observations
Device calibration outputs
Routine phenotyping
Detect laboratory-specific effects
Support historical controls
Refine endpoints
Train AI systems
Reveal patterns invisible in single studies
Unstructured, shadow data are a landfill. Structured, they become an ethical multiplier — one animal’s data informing many future analyses.
The stakes, made concrete
Virtual control groups use historical control data and modelling to reduce or replace concurrent control animals — an active area in toxicology and regulatory science (e.g. VICT3R). The promise is real: fewer control animals, better reuse, more efficient designs.
Virtual control groups require structured, high-quality, well-described datasets. Every poorly annotated control animal from the past is a missed opportunity for Reduction in the future — an animal-welfare opportunity that expired when context was lost.
The culture problem
Paying the debt
Before
Minimal metadata set
Defined before the experiment begins; captured automatically from devices, ELNs and pipelines.
During
Ontologies + provenance
Controlled vocabularies for animals, procedures, devices, behaviours; raw data preserved with provenance.
Publish
Negatives + federation
Register negative and neutral findings; federate when data cannot be openly centralized.
Govern
Stewards + incentives
Data stewards as collaborators; data quality embedded in ethics, peer review and evaluation.
The logical extension of the 3Rs into the data age. Data stewardship is part of the ethical contract.
Damien Huzard, PhD · Neuronautix · 2026-06-19
neuronautix.com/contact