FRIDGE: Federated Research Infrastructure by Data Governance Extension

FRIDGE is a “satellite TRE” approach for safely using supercomputers as part of a Trusted Research Environment (TRE). This extends the information governance and data security controls of an existing TRE to an isolated part of a supercomputer, allowing it to be safely used for research with sensitive data. FRIDGE takes a shared responsibility model, defining what the supercomputer administrators need to do to isolate a part of their system as a secure “TRE tenancy”. Control of this tenancy is then handed over to the TRE operator, who deploys components of their TRE within it and manages these under the same controls as the rest of their TRE. Researchers using the TRE are then able to safely use the supercomputer’s specialist computer processors within this tenancy for their research with sensitive data.

Enabling the safe use of supercomputers for sensitive data is important because we have a large wealth of health data in the UK, and artificial intelligence (AI) has the potential to enable research using this data that could have a huge positive impact improving the quality of our health and healthcare treatment. However, this data is highly sensitive and can therefore only be used for research under strong data security and information governance controls. The government has made a big investment in large AI supercomputers such as AIRR (the AI Research Resource), which can be used for AI-driven research on large datasets. However, these supercomputers were not originally built to provide the level of data security and information governance required for research with sensitive data. Enabling safe research on sensitive data using these AI supercomputers has the potential to give insights that can improve all our health and wellbeing.

In this DARE UK Early Adopter project, we have shown that the FRIDGE satellite TRE approach works by extending the information governance and data security controls of the Alan Turing Institute’s TRE to securely isolated parts of the two AI Research Resource supercomputers. There is still significant work to do to further develop FRIDGE from this initial proof-of-concept demonstration into something that can be deployed at scale in production. However, we have shown that both the technical approach and the shared responsibility information governance model work in a realistic setting across real-world TREs and supercomputers.

This end of project report provides an overview of the FRIDGE approach and reflections on the project and the wider DARE UK programme it was part of.