The Catalogue That Doesn’t Exist Yet: What Europe’s Cancer Mission Asks Consortia to Promise
On 15 September at 17:00 Brussels time, the 2026 Cancer Mission calls close. Seven topics, €124.19 million, eighteen grants expected across the whole call. The first topic on the list, Virtual Human Twin models for cancer research, puts €35 million on the table for four projects. It also contains one sentence that consortia are signing up to without much discussion.
Here it is, from the topic text: newly collected patient datasets must be described with metadata records in the EU dataset catalogue of the European Health Data Space.
Three clocks are running behind that sentence, and none of them is set to 15 September. The rules that will say what a metadata record has to contain are due in March 2027. The chapter of the Regulation that makes the description a binding duty applies from March 2029. The proposal is due in a fortnight.
Fifteen days and a single stage
HORIZON-MISS-2026-02-CANCER-01 opened on 10 February and closes on 15 September. It is a Research and Innovation Action with an expected EU contribution of €8 to €9 million per project against a topic budget of €35 million, and four grants indicated. Thresholds are 4 for Excellence, 4 for Impact and 4 for Implementation, with a cumulative threshold of 12.
Single stage matters here. There is no short outline to test the water, no evaluator feedback before committing. Whatever a consortium writes about data governance in September is what gets scored, and there is no second draft. The portal currently lists 125 partner search announcements on this topic alone, which is a reasonable proxy for how many teams are circling four grants.
The topic asks for four things: dynamic multiscale virtual human twins integrating AI over FAIR multimodal data; validation against longitudinal patient data with evidence of clinically meaningful observations; enrichment of the UNCAN.eu research platform and the Advanced VHT Platform under open science principles; and cross-disciplinary collaboration involving clinicians and patients throughout.
The third of those carries the sentence about the EHDS catalogue. It reads like an administrative footnote. It is the one that commits a consortium to an infrastructure that is still being built.
Read the third bullet
Article 77 of the European Health Data Space Regulation is titled “Dataset description and dataset catalogue”. It requires health data access bodies to publish a machine-readable catalogue describing available datasets in metadata form, covering source, scope, main characteristics, nature of the data and the conditions for access. National catalogues must be available in at least one official language. The catalogue provided by the Union health data access service must be available in all of them.
Then comes paragraph 4, and it is worth quoting closely. By 26 March 2027, the Commission shall, by means of implementing acts, set out the minimum elements health data holders are to provide for datasets and the characteristics of those elements.
So the shape of the thing is fixed and the content is not. What a description must cover is a deliverable the Commission owes itself, with a deadline roughly six months after this call closes. A consortium submitting in September 2026 is therefore committing to a technical standard whose content will be settled after the grant agreement is signed.
What the EHDS actually asks, and when
The dates are worth getting right, because the summaries circulating in the sector do not always agree, and more than one places secondary use in the wrong chapter.
Regulation (EU) 2025/327 entered into force on 26 March 2025 and applies from 26 March 2027. Chapter IV, which covers secondary use and runs from Article 50 to Article 81, applies from 26 March 2029. Some data categories under Article 51 follow later still, from 26 March 2031: genomic and other molecular data, clinical trial data, research cohorts and survey data among them, which is to say much of what a virtual human twin project would actually want.
Article 77 sits inside Chapter IV. The duty to describe datasets in the catalogue is not yet a duty anyone can be held to, and will not be one until 2029. That does not make the call requirement decorative. It makes it anticipatory, which is a different problem and arguably a harder one to write a work package against.
There is a second provision consortia should read now. Article 78 creates a Union data quality and utility label. Paragraph 1 makes it optional for datasets generally. Paragraph 2 does not: datasets with electronic health data collected and processed with the support of Union or national public funding shall have a data quality and utility label. A Horizon Europe grant is Union public funding. The label covers data documentation, metadata, the data dictionary, formats and standards, completeness and accuracy, the maturity of quality management processes including bias examination, population coverage, and more. Its visual characteristics and technical specifications are due from the Commission by 26 March 2027, on the same clock as the metadata elements.
Why the sentence is in the call at all
It would be easy to read all this as an oversight, a requirement drafted for a world that has not arrived. It is not. It is Article 85.
That article requires contracting authorities and Union institutions to reference the applicable technical specifications and standards of the Regulation when formulating tender documents or calls for proposals, and when defining the conditions for Union funding. Paragraph 2 goes further: the criteria for obtaining funding from the Union shall take into account the requirements developed in the framework of Chapters II, III and IV.
The Regulation, in other words, instructs the Commission to put EHDS requirements into calls for proposals. The sentence in the Cancer Mission topic is that instruction working as designed. Which means it will not be the last one, and consortia that treat it as boilerplate this year will meet it again, more explicitly, in the calls that follow.
The AI Act sits underneath
Most discussions of AI and the EHDS reach for the high-risk regime, and it does apply: a virtual human twin that stays a research instrument is one thing, and a virtual human twin that becomes decision support for diagnosis or treatment is a different object, carrying obligations on risk management, data governance and human oversight. Nothing here is a legal opinion, and no specific product can be classified from the outside.
But there is a more concrete link, and it is written into the text rather than inferred. Article 78(6) provides that the implementing acts specifying the quality and utility label shall take into account the requirements in Article 10 of Regulation (EU) 2024/1689, the AI Act’s data governance article, together with any common specifications or harmonised standards supporting them.
That is a direct bridge. The label a publicly funded dataset must carry is being specified with reference to the AI Act’s expectations for training data. A team building its data pipeline for the label is doing work that counts twice, and a team that treats the two frameworks as separate compliance exercises will do it twice instead.
Where proposals are actually won
Excellence is one criterion of three, and a strong scientific case clears the threshold without winning the grant. In a topic where four projects will be funded out of a field suggested by 125 partner searches, separation happens in Implementation, and Implementation on this topic means data.
Which is oddly good news for anyone reading this in the last two weeks before the deadline, because the moving specification is not a problem to hide. It is a chance to be more precise than the competition.
Three questions worth putting to a consortium before 15 September.
What does the data management plan actually commit to? “We will comply with the EHDS” is a sentence an evaluator cannot score. A plan that names Article 77, states that the minimum metadata elements are due by March 2027 and that Chapter IV applies from March 2029, and describes how the project will adopt the elements when they land, is a plan that shows the team has read the file.
Who owns the metadata work, and is it budgeted? Describing datasets to a machine-readable standard, keeping descriptions current across a multi-site collection, and carrying a quality and utility label under Article 78 is real effort. If no work package holds it and no person-months are attached, the commitment is decorative.
What happens if the specification lands differently than expected? Consortia that build their pipeline around a single assumed schema inherit a rework risk in year two. Consortia that separate their internal data model from the export layer do not.
None of this replaces the science. It decides which good science gets funded.
Sources
- European Commission, EU Funding & Tenders Portal, topic HORIZON-MISS-2026-02-CANCER-01, Virtual Human Twin (VHT) Models for Cancer Research: topic text, mandatory activities, thresholds, budget overview for all seven topics, partner search count. Read directly on the portal, 31 August 2026.
- HaDEA, Cancer Mission calls 2026: call status, publication and opening dates, deadline, list of the seven topics. Verified 31 August 2026.
- Regulation (EU) 2025/327 on the European Health Data Space: Article 77 (dataset description and dataset catalogue), Article 78 (data quality and utility label), Article 85 (additional requirements for public procurement and Union funding), Article 105 (deferred application). Read on EUR-Lex, 31 August 2026.
- Bird & Bird, Health Data Holders’ obligations under the EHDS (31 March 2025): application dates for Chapter IV duties, the 2031 data categories, status of the Article 77(4) implementing act.
- Osborne Clarke, European Health Data Space Regulation and TEHDAS2 lay down a new roadmap (24 February 2026): HealthDCAT-AP as the proposed metadata model, TEHDAS2 consultation status.
For how OneSynergy works on EU-funded research projects — from portfolio positioning to consortium roles — see the EU Research and Innovation Funding service page.
