The Framework

The workshop produced a co-designed framework for coordinating national biomedical data science resources — spanning infrastructure, training, and people — structured around five interdependent pillars. This framework was agreed in principle by all participants and provides the foundation for a phased implementation roadmap. It reflects the community’s conviction that infrastructure alone is insufficient: without coordinated training, sustainable career pathways, and shared expertise, even the best-resourced platforms will fail to deliver their potential.

How this framework was built It emerged from structured discussions involving researchers, infrastructure providers, training specialists, funders, and institutional leaders — all responding to evidence gathered over the preceding year across infrastructure provision, training availability, workforce capacity, and institutional costs. It reflects what the community identified as essential rather than any single organisation’s needs.

The five pillars

The framework rests on five pillars, each addressing a distinct dimension of the coordination challenge. Three pillars address the enabling conditions (infrastructure, governance, authentication and access), while two address the often-overlooked human dimensions (data ownership and IP, careers and training). Importantly, participants recognised that these pillars are highly interconnected — progress on one often depends on progress on others — and that the people-focused pillars are not secondary to the technical ones.

Pillar 1: Infrastructure

What this covers

The physical and digital infrastructure required for biomedical data science: compute (including HPC and cloud), storage, software platforms, and the skilled people who operate them.

What already exists

Most nations already have significant infrastructure assets — HPC centres, cloud partnerships, domain-specific platforms, and emerging national data initiatives. The problem is the current fragmentation: resources exist but are disconnected, duplicated, or inaccessible to researchers who need them.

Key findings from the workshop

Storage and compute were identified as essential phase-1 requirements alongside file synchronisation and sharing capabilities
Cloud access through national HPC providers was seen as an anchor for the model
Existing platforms (national data repositories, domain-specific tools) should be connected, not replaced
A skills shortage in research software engineering and data engineering was identified as a barrier as significant as any hardware gap

What can be delivered in 12-18 months

Exemplar projects demonstrating cross-institutional workflows
A landscaping exercise documenting what exists and where gaps lie
Community webinars to build awareness and connection

What still needs scoping

Full inventory of institutional infrastructure across participating organisations
Costing models for shared services vs. duplicated institutional provision
Gap analysis against international comparators
Solid worked examples showing end-to-end workflows

Pillar 2: Governance and decision rights

What this covers

How decisions are made about data access, infrastructure use, resource allocation, and strategic direction. This includes governance of the shared model itself and the governance frameworks applied to data within it.

Key principle identified

Governance should be based on principles rather than project-based funding. The question is how to evaluate risk under this approach.

Key findings from the workshop

Some examples of good governance practice already exist and should be extended rather than reinvented
A modular approach was suggested — described as “3-in-1 Lego sets for governance” — flexible enough to accommodate different data types and use cases
University responsibility over data generated, combined with a collaborative approach to access
Data curation recognised as a form of “currency” within the system

What can be delivered in 12-18 months

Landscape and discovery exercise mapping existing governance models
Identification of pathways and access model combinations
Clarity on the funder-university relationship in data governance

Risks of inaction

National research capability falls behind international comparators
Health inequalities widen without data-driven insights
Research expertise leaves for better-resourced environments
Further divergence between health and biomedical data ecosystems
Universities continue to duplicate governance overhead individually, adding cost without improving outcomes

Pillar 3: Authentication and access

What this covers

How researchers, collaborators, and industry partners are identified, verified, and granted appropriate access to shared infrastructure and data. This includes the concept of a “research passport” — a portable identity that works across institutional boundaries.

What already exists

Institutional single sign-on systems and federated identity (e.g. eduroam-style)
National researcher identifiers (e.g. ORCID)
Government identity systems
Existing safe-haven and trusted research environment access processes

Key findings from the workshop

A research passport concept emerged — a portable credential not tied to a single host institution
An existing national safe researcher registry was cited as a model to build on
Industry access must be designed in from the start, not bolted on later
Current authentication systems are fragmented: what works at one institution may not be recognised at another

Concrete next steps identified

Develop a working group specifically for authentication
Compile a list of common requirements for a minimum viable product
Formalise and document requirements for a biomedical research identifier
Identify exemplars of what is and is not working well currently
Engage with data controllers, institutions, industry, and government
Engage with national data bodies to promote shared requirements

Risk of inaction

Too many disconnected systems create liability for bad actors and barriers for legitimate researchers.

Pillar 4: Data ownership and intellectual property

What this covers

Who owns research data, how IP is managed in collaborative and shared environments, and how the rights and interests of data subjects (ultimately, citizens) are protected and respected.

What already exists

Domain expertise embedded in research groups
Research data management teams working with contracts offices
Shared service legal models at sector level
Creative Commons and similar licensing frameworks
PhD-level data agreements and institutional IP policies

Key findings from the workshop

Agreement from all participating institutions to start conversations around a shared services model for legal and IP support was identified as a deliverable within 12-18 months
A common resource pool for IP (including IP training for academics and professional services staff) was proposed
Industry terms need further exploration — they should be treated equitably but may require different models
A pragmatic “80/20 rule” was explicitly endorsed: the framework may never fit every scenario, and that is acceptable

A powerful framing that emerged

How do we ensure there is a return to the “real” data owners — that is, the nation’s citizens?

This question reframes data infrastructure as a public good and has potential as a narrative for public engagement and government advocacy.

What still needs scoping

Understanding of what legal agreements and frameworks currently exist across institutions
Industry requirements and acceptable terms for data access and IP sharing
Non-cost models that allow mutual benefit between academia and industry

Pillar 5: Careers and training

What this covers

Career pathways, professional development, training provision, and recognition for the people who make data science infrastructure work: data scientists, research software engineers, data stewards, bioinformaticians, and technical professionals. The survey evidence was stark: 76% of respondents working with mission-critical biomedical data had no formal training in data science, and existing training resources — where they existed — were fragmented across institutions and invisible to those who needed them most.

Key findings from the workshop

The lack of career recognition and sustainable pathways for data professionals was identified as both a barrier to infrastructure development and a failure risk for the entire initiative.

Consistent signposting to existing reputable training resources (e.g. Software Carpentry, Data Carpentry) was seen as a quick win
A scoping exercise on current career development frameworks and accreditation schemes was identified as a priority
Cross-disciplinary mentoring schemes were proposed to connect data scientists with domain researchers
The Technician Commitment was cited as a recognition model to learn from

What can be delivered in 12-18 months

Community network events connecting PhDs, data scientists, and non-data-science mentors
An inventory of what training and career development currently exists
Scoping of accreditation frameworks and their applicability

What still needs evidence

Comprehensive mapping of available training and career pathways
Assessment of the skills shortage (data science, data engineering, AI/ML, ethics, IT security)
Understanding of what career development opportunities are actually taken up vs. merely offered

Cross-cutting themes

Three themes emerged that cut across all five pillars:

People before platforms Career pathways, skills shortages, training fragmentation, and the need for dedicated coordination capacity were common discussion points. Three-quarters of biomedical researchers work with data they consider mission-critical but lack formal training in the methods they use daily. Any implementation plan that leads with technology and buries the people and training dimensions will fail to deliver the workforce needed to make shared infrastructure effective.

Integration, not re-invention Good training resources already exist but are fragmented and hard to find; infrastructure assets are significant but poorly coordinated. The framework should be “designed to be connected” — linking existing institutional investments, national data programmes, training provision, and international standards bodies into a coherent, discoverable whole.

Sustainability from day one Short-term or project-based funding was the single most cited failure risk. The framework must include a sustainability model — financial, institutional, and operational — from the outset. Baseline funding (not just pump-priming) was the clear preference.

Implementation roadmap

The workshop produced a phased roadmap for taking the framework from agreement to action:

Phase	Timeline	Key Activities
Stakeholder mapping	1-3 months	Identify all relevant parties; convene a working group; agree terms of reference; produce a glossary of shared terms
Landscape review	3-6 months	Audit existing infrastructure, governance, and training across institutions; identify gaps and duplication; develop a business case
Analysis and pitch	6-12 months	SWOT/gap analysis; government and funder engagement; formal strategic case for investment
Implementation	12+ months	Defined deliverables including shared resources, datasets, and IP frameworks; ongoing coordination via task-and-finish groups

A caution from the workshop Participants explicitly warned about scope creep. A note on one flipchart read simply: “SCOPE CREEP” with arrows pointing to the need for focus. The 80/20 rule applies — the framework will never accommodate every scenario, and that is acceptable. The mechanism for maintaining focus must be designed into the governance of the working group.

Next: The Toolkit — a practical, phase-by-phase guide for running similar initiatives