WellFAIR webinar recap — FAIR data, the 3Rs, and the path to AI/VCG

What we covered

The talk opened with the WellFAIR argument: data are the only enduring outcome of an animal experiment, so if those data are lost, under-described, or impossible to reuse, the ethical justification for the original animal use weakens [1]. FAIR is therefore not just a compliance overlay — it is ethical infrastructure for Reduction (avoiding unnecessary duplication), Refinement (welfare-relevant metadata captured at the moment of decision), and the slow path to Replacement through reusable evidence [1].

We anchored the community context in the COST TEATIME long-read, which makes the case that better behavioural data culture and shared infrastructure are pre-conditions, not nice-to-haves [2].

FAIR3R.fr — the French CNRS framework

Benoit demoed FAIR3R.fr, the French CNRS-supported initiative that ties FAIR data practice explicitly to the 3Rs. The point we kept returning to: institutional frameworks like FAIR3R make the ethics argument concrete — they push metadata, sharing, and reuse into the daily workflow rather than leaving them as a manuscript-stage afterthought.

Metadatapp — born-FAIR capture in practice

I demoed Metadatapp, our born-FAIR capture tool. The design idea is simple: capture experimental context as structured, ontology-aware metadata at the moment of decision — protocol design, cohort registration, instrument setup — so that ARRIVE-style reporting and downstream reuse fall out of the workflow rather than being reconstructed at manuscript stage [1]. For the HCM-specific context — sensor modalities, housing parameters, deep behavioural metadata — we pointed to the Open Access COST TEATIME book on Home-Cage Monitoring [3].

AI agents and the path to VCGs

We closed on AI agents and virtual control groups. The argument: agents that read, validate, and synthesise preclinical evidence become genuinely useful only when the substrate is FAIR, ontology-anchored, and provenance-complete. VCGs are the most concrete near-term payoff — historical control data is only reusable if its metadata is good enough to test comparability [1].

Audience Q&A — what came up

Four questions stood out in the chat and the live Q&A. They map cleanly onto the gaps that any FAIR-for-3Rs programme has to address.

• RRIDs in repositories (Esther Pearl). Does the repository have fields for RRIDs — antibodies, software tools, model organisms? Short answer: yes, and they should be mandatory at the minimal layer. Persistent identifiers for reagents and tools are one of the cheapest interoperability wins available, and they make downstream AI-assisted curation tractable.
• Protecting data from AI (Lidija Križančić Bombek). Can the data be protected so AI cannot see it? FAIR is not synonymous with open — the "A" in FAIR is Accessible under defined conditions. Access tiers, licences, embargoes, and machine-readable usage terms all belong in the same metadata layer. The realistic answer for sensitive data is structured access control, not opacity.
• Data steward vs FAIR culture (Leonardo Restivo). For a new PI: hire a data steward, or grow a FAIR culture so every lab member is autonomous? The honest answer is both. A culture without a steward erodes within one PhD cycle; a steward without a culture becomes a bottleneck. In small labs the steward role can be partial or shared, but somebody has to own the schema, the validation, and the onboarding.
• Repo vs OSF (Raphaelle Bidgood). How does the repository differ from OSF? OSF is a general-purpose project hub — strong for openness and project structure, weak on domain-specific schemas. The FAIR3R/Metadatapp direction is ontology-anchored, domain-aware capture: structured fields for species, housing, sensor, procedure, with validation against controlled vocabularies. The two are complementary rather than competing.

Resources shared in the chat

• COST TEATIME long-read — community-level case for behaviour, data, and the 3Rs [2]
• WellFAIR paper — Petit-Demoulière & Huzard, Neuroscience Applied 2026 [1]
• TEATIME Behaviour Forum — discussion venue for animal behaviour methodology [4]
• Open Access HCM book — Springer, COST TEATIME [3]
• Webinar replay — full recording on Google Drive
• COST TEATIME YouTube — official replay will also appear here
• Original .pptx slides on Zenodo — doi.org/10.5281/zenodo.20612830