Research use only

Onkydra is an in-silico stratification workspace. It takes a target and an indication, samples a simulated cohort from a Gaussian copula joint fitted to real anchor evidence, runs a ridge-plus- foundation-model ensemble over that cohort, argues with itself where the two models disagree, and hands back a citation-backed narrative that a working scientist can take to the next meeting.

It is a tool for generating and prioritising hypotheses about drug programmes in rare cancer. It gives you candidate strata ranked by an uncalibrated ordering signal, ranked resistance pathways, cohort frequencies bootstrapped over the real anchor cohort rather than over the simulated draws, and a declared proxy status for every evidence layer.

Onkydra is not a medical device. It is not an in-vitro diagnostic. It is not a treatment recommender. It is not a clinical decision-support system. It is not a substitute for wet-lab, animal-model, or clinical validation. It cannot be used to make care decisions for a specific patient, and it cannot be used to support a regulatory submission without independent human expert review and independent validation.

Under Regulation (EU) 2017/745 (the Medical Device Regulation) and Regulation (EU) 2017/746 (the In Vitro Diagnostic Regulation), software qualifies as a medical device or an in-vitro diagnostic only where its intended purpose falls inside the scope of those regulations. The MDCG 2019-11 guidance frames this using the intended purpose declared by the manufacturer. Onkydra's intended purpose is research use only: hypothesis generation and target prioritisation for drug developers. We do not intend it for the diagnosis, prevention, monitoring, prediction, prognosis, treatment, alleviation, or management of disease in any identified or identifiable patient. Accordingly, Onkydra is not offered as a medical device or as an in-vitro diagnostic medical device, has not been CE-marked under either regulation, and has not been cleared or approved by the U.S. Food and Drug Administration.

The workspace does not ingest patient-identifiable data. Cohorts are statistical draws from public distributions, never records of real people. If you upload data that you do not have the right to process, or that would identify a real patient, you are in breach of the Terms of service.

Every claim in every report is backed by a citation. Every confidence interval is resampled from a real anchor cohort (n=60 for DMG), not from the 1,000 posterior draws. Every disagreement between the ridge baseline and the foundation model is surfaced by the Honesty Critic rather than hidden. That is the boundary of what the workspace can honestly claim.

Beyond that boundary, you are on your own. A prioritised target needs a wet-lab experiment. A promising biomarker needs an IHC, FISH, qPCR, or NGS assay run on real specimens. A candidate trial design needs an IRB, a sponsor, and a regulatory submission. Onkydra brings you to the edge of the wet-lab budget conversation. It does not replace the conversation.

By using Onkydra you accept the following. You will not present Onkydra outputs as diagnostic findings, clinical recommendations, or the basis of an individual patient-care decision. You will not submit them to a regulator as pivotal evidence without independent validation. You will apply qualified scientific and clinical judgement to every output before acting on it. You will treat the Honesty Critic's flagged uncertainties as first-class findings rather than footnotes. And you will cite Onkydra clearly, along with the underlying public datasets, in any downstream publication or memo.

If a use case might sit close to the medical-device boundary, contact us before you deploy it. We would rather have that conversation than see a good research tool used for the wrong purpose.

The research-use-only banner appears on the header and the footer of every report the workspace generates, and is embedded in the JSON envelope of every artefact export. The stable disclaimer identifier is pinned in the audit log so a downstream auditor can verify exactly which wording your report was generated against. Outputs are labelled as AI-generated in line with the transparency approach set out in Article 50 of the EU AI Act (Regulation (EU) 2024/1689).

A worked example of the disclaimer in context is on the methodology page, which walks through a full DMG run end to end. A short-form version answering the “what does this mean?” objection is in the FAQ.

Questions or clarifications on the scope of the Service, or on a specific report, should go to hello@onkydra.com. We reply the same working day.

This page is a plain-language template and will be reviewed by qualified Irish counsel before Onkydra transacts a regulated engagement. It does not constitute legal advice.