At the Francis Crick Institute, CIO James Fleming describes IT not as a back-office service but as part of the infrastructure that lets research teams work with complex data, collaborate securely and test scientific ideas. His account traces the shift from high-performance computing (HPC) to data science embedded in laboratory work, and explains why trusted research environments (TREs) can help collaborators analyze sensitive data without treating access as an all-or-nothing choice.
The details below come from Computer Weekly interviews published on 30 October 2024 and 3 October 2022. Figures, timelines, supplier relationships and descriptions of systems are attributed to Fleming or the interviews; they are not independently audited institutional metrics.
How Fleming connects IT leadership to scientific work
Fleming joined the Crick in October 2018 after 11 years at BT, where he worked on telecommunications projects including the UK fibre broadband rollout. He told Computer Weekly in 2024 that the institute had more than 2,000 staff and students and over 100 research groups. Its scale and scientific mission shape the challenge: researchers need computing and data services that can support varied work rather than a single, fixed workflow.
Fleming’s guiding question is practical: “how do I work with my dataset?” The institute’s aim, as he described it, is to enable world-class science. That means bringing technical capacity close to the research question—whether a team needs intensive computation, a way to analyze data, or a controlled means of working with external collaborators.
#1 Best Overall
From high-performance computing to data science in the lab
Fleming described the Crick’s computing capability as evolving from an early investment in HPC to a refreshed second generation. In parallel, data science has become part of routine laboratory work. These capabilities serve related but distinct needs: HPC provides substantial computing capacity for demanding workloads, while data-science practices help researchers interrogate and interpret datasets as part of their scientific work.
The institute brought scientific computing together with the rest of IT, a change that Fleming said led to creation of the CIO role in 2021. In the 2024 interview, he described an open-source software stack running on hardware supplied by NVIDIA for GPUs, Dell for CPUs, and Lenovo and IBM for storage. He also identified Snowflake as a key provider for the institute’s trusted research environments. These are reported supplier relationships, not recommendations of particular products or a blueprint that every research institution should adopt.
What the COVID-19 response showed about adaptable infrastructure
Fleming cited the Crick’s response to the COVID-19 pandemic as an example of repurposing institutional infrastructure. In 2020, he said, the institute adapted its systems for a testing pipeline supporting acute trusts in London. He reported that the pipeline was operating in nine days, with a staff-testing solution following in a further nine days. Those are timings Fleming gave in the 2024 Computer Weekly interview; they describe that reported effort, not a general delivery benchmark.
The broader point in his account is that infrastructure has to be adaptable enough to support work beyond its original plan. In research, that can mean enabling a new collaboration or accommodating unfamiliar data—not merely increasing compute capacity.
How trusted research environments enable collaboration on sensitive data
A TRE offers a controlled environment in which approved users can analyze data under defined conditions. In Fleming’s description, TREs complement rather than replace HPC: one provides a managed setting for collaboration and data access, while HPC addresses compute-intensive work. The 2024 interview reported a dozen TRE projects running concurrently at the Crick at that time.
This approach speaks to a common research problem: collaborators may need to work together on sensitive datasets, but access must reflect the data’s sensitivity, permitted uses and governance requirements. A controlled shared environment can make it possible to grant useful analytical access without assuming that every collaborator should receive an unrestricted copy of the underlying data. The exact controls depend on the project; the interviews do not establish a universal compliance outcome for TREs.
In a 2022 interview about DARE UK work on a repeatable framework for on-demand TREs, Fleming described an architecture using Snowflake for data sources, Apache Airflow for workflow, DBeaver for data extraction, transformation and loading, Okta for authentication, and ServiceNow for additional controls. He said a secure, auditable environment could be built in around 30 minutes. That is Fleming’s description of the reported approach in 2022, not a general performance guarantee or a claim that the same components and partnerships remain current.
The 2022 article named BT, the Institute of Cancer Research and the Rosalind Franklin Institute as consortium partners, with Infinite Lambda and Snowflake as technical partners. Fleming said four clinical research projects used the framework, spanning cancer, Parkinson’s disease, hepatitis B and schizophrenia.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →How data access decisions depend on the project
Fleming’s 2022 account makes clear that access design depends on more than the technology. Decisions turn on how well a dataset is understood, what level of control is needed, the data’s sensitivity and applicable jurisdictional requirements, and the practical form of collaboration. A team may need direct file access, programmatic access or a managed shared environment; the appropriate choice depends on the approved work and its governance.
Rank #4
- Dataset maturity: Is the dataset’s structure and meaning sufficiently understood for the planned analysis?
- Access granularity and auditability: What should each participant be able to do, and what activity needs to be recorded?
- Sensitivity and jurisdiction: Which ethics, legal, cross-jurisdiction or data-sovereignty constraints apply?
- Collaboration model: Do researchers need files, programmatic access or a controlled environment in which they can work together?
- Accountability: How will roles, responsibilities, ethics approvals and collaboration agreements be reflected in the technical workflow?
The interviews describe considerations, not a complete implementation checklist. Institutions must determine the applicable controls for each dataset and collaboration rather than infer that a particular platform alone settles governance.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Why provenance matters when researchers use AI
Fleming said AI and machine learning have long been used in scientific work, but he emphasized that applying them rigorously requires more than generating an output. Researchers need to know where the data came from, how a model’s performance changed over time, and what an analysis contributes to understanding a disease mechanism. As he put it, “You’ve got to be rigorous with the provenance of what you’re working with, the data you’re using, how you’ve improved the model’s performance over time, what that analysis has done for your mechanistic understanding of the disease, and so on.”
That is a case for traceable, scientifically meaningful analysis—not a claim that the Crick rejects generative AI or that AI results are automatically reliable. Provenance, performance tracking and interpretation remain part of the work needed to judge what an analysis can support.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Best Value
The next challenge: connecting clinical, environmental and geospatial data
In the 2024 interview, Fleming identified integration of clinical research with environmental and geospatial data, sensors and other nontraditional sources as a next challenge. He explicitly said the solution was not yet known. His question—“How do we bring that data and connect it to a clinical lab?”—captures a problem that is both technical and scientific: linking different forms of information in ways that preserve their context and support meaningful analysis.
For research IT, his account suggests that success is not simply a matter of buying more compute or selecting a single data platform. It involves adaptable infrastructure, appropriately governed collaboration and the ability to understand where data came from and what conclusions analysis can genuinely support.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
