AI and the future of immunology

Artificial intelligence (AI) is already beginning to change how immunologists work, from analysing complex datasets and supporting literature searches to exploring new approaches to diagnosis, drug discovery and research. 

As these technologies continue to evolve, what opportunities could they offer immunology, and what challenges do researchers and clinicians need to consider?

In this feature, Dr Zania Stamataki and Dr Jason Cosgrove explore how a new generation of AI tools could transform the way immunologists analyse and interpret complex biological data, while our AI expert considers what these changes could mean for researchers in practice.

The autonomous bioinformatician – balancing the promise and pitfalls of emerging AI technologies

Dr Zania Stamataki, Associate Professor of Viral Immunology at the University of Birmingham and member of the Immunology News Editorial Advisory Board

Step into almost any immunology lab today, and you will find researchers using AI to polish an abstract or troubleshoot a stubborn line of R code. Behind the scenes, however, a much more profound shift is occurring. We are moving rapidly from AI "chatbots" that respond to individual prompts to agentic AI, systems designed to act as autonomous computational partners.

For immunologists, this transition represents both an incredible opportunity to democratise data analysis and a new set of practical challenges.

Beyond the chatbot: what is an agent?

Traditional AI requires you to guide it step-by-step. An AI agent, by contrast, is given a high-level goal. For example: "Take this raw single-cell RNA-seq dataset from the Gene Expression Omnibus (GEO), filter out the low-quality cells, perform batch correction, and identify the T-cell subsets."

The agent then writes the script, runs it, reads the terminal errors, alters its code, and loops recursively until the job is done. 

For a non-computational bench immunologist, this feels like magic. Having spent the last few months putting Google DeepMind’s specialised scientific agent packages through their paces, I can attest that the speed is breathtaking. These tools effectively bypass the traditional bioinformatics bottleneck, allowing researchers to interrogate heavy multi-omic datasets with unprecedented ease. At the same time, they challenge us to critically think how computational research is conducted and how scientific decisions are made.

The pitfalls lurking in the loops – the challenges of agentic AI

Delegating the scientific workflow to an autonomous system introduces subtle, compounding risks that our community must evaluate:

• Silent data biases: When an agent encounters an error or a noisy dataset, it may make small choices to get the analysis running. If it adjusts a normalisation threshold or cell-filtering metric without human oversight, it can subtly alter downstream clustering, potentially "hallucinating" rare immune cell states or masking vital biological signals
• The university "sandbox" gap: To run code safely, AI agents require isolated digital environments called sandboxes. While commercial corporate labs can deploy these instantly, our risk-averse university IT departments are understandably lagging behind due to cybersecurity concerns. This leaves academic labs unable to officially deploy the very tools that could accelerate their research at pace with industry
• The "token tax" on public grants: Agentic AI systems can also be expensive to run. Unlike traditional software licences, costs are often based on usage. An agent caught in repeated coding or troubleshooting loops can consume large numbers of API tokens, making expenditure difficult to predict. For academic labs operating on rigid, fixed-budget public grants, this stochastic spending is difficult to budget for.

Keeping the immunologist in the loop

So, how do we move forward? The answer isn’t to reject automation, but to adopt it responsibly. 

As a community, we need to establish standard "human-in-the-loop" checkpoints. An agent can do the heavy lifting of writing the pipeline, but a qualified immunologist must still sign off on the critical biological transitions, such as post-normalisation distributions and cell-type annotations.

Agentic AI has the potential to accelerate immunological discovery, liberating researchers from the command line and allowing them to spend more time asking biological questions. However, as these tools become more capable, scientific judgement becomes even more valuable. To safeguard the integrity of our research, we must ensure that while AI powers the analysis, the immunologist keeps a firm hand on the wheel, validating the results, applying scientific judgement and taking ultimate responsibility for the conclusions.

Putting AI into practice

Dr Jason Cosgrove, Principal Scientist, Computational Medicine

How can immunologists make the most of AI while using it responsibly?

AI is a new and rapidly evolving technology, and as such, there are not yet any clear consensus guidelines or principles on how best to apply it to immunology research. However, there are three key things that immunologists can do: 

1. Train and upskill in this area through workshops, taught practical sessions, or online courses
2. Actively experiment with different tools to see how best to apply them to their own research goals
3. Create communities such as AI working groups at a departmental or institutional level to share experiences and collectively identify strengths, limitations, and areas requiring caution.

Responsible use also requires clear procedures, particularly around patient data.  Not all AI tools are secure or GDPR-compliant, so research teams need to establish which models to use and when, with explicit guidelines on data access, use, and protection.

How do you think AI will change the day-to-day role of immunology researchers over the next five years?

I believe researchers will increasingly collaborate with AI assistants to support literature reviews, hypothesis generation, experimental design, and other routine tasks. As a result, immunologists must learn to think not only as scientists but also as engineers – developing the skills to construct, test and validate robust AI pipelines and workflows.

AI will also reduce the barrier to entry for data analysis, enabling more researchers to make use of publicly available datasets for grant writing and preliminary insights before projects even begin. In the longer term, AI-driven laboratory automation and predictive modelling should make immunology research more resource-efficient, improve reproducibility, and help refine the number of animals used in experiments.

Should researchers be worried about AI replacing research jobs?

My personal view is that AI will fundamentally transform the day-to-day role of immunology researchers in the coming years. Critically evaluating AI-generated outputs will become a core skill, requiring experience and expertise in experimental design, data analysis and computer programming. Despite such changes, scientific judgement and rigorous application of the scientific method will be more important than ever to ensure that results are reliable and reproducible.

One area, however, where I think we need good policy decisions is on improving access to entry-level positions. With routine tasks increasingly automated, there's a risk of fewer opportunities for those starting out. I would like to see more well-paid positions available to those coming directly out of a degree or master’s programme, with clear progression schemes that help them develop without necessarily needing to do a PhD.

What's one skill every immunologist should start developing today?

Context engineering! You will probably have noticed already that if you don't give an LLM a detailed prompt, you won't get a good answer. Using the principles of context engineering, you can turn your LLM into an AI assistant by putting highly detailed prompts into some input files so they can be reused again and again. To use a research lab as an analogy, it is equivalent to providing team members with a set of protocols and reagent lists: the key background knowledge and organisational structure needed to get high-quality research outputs.

Ready to build your AI and bioinformatics skills? Explore BSI training designed to help researchers use these tools confidently and responsibly

AI training for immunologists and the biosciences

Join live, online sessions exploring how AI can support research and how to use these tools effectively and responsibly.

Bioinformatics and R training for wet-lab biologists 

Build your skills at your own pace with flexible self-paced learning, or join live training designed to help wet-lab researchers gain confidence analysing biological data.