Sometime between 2006 and 2010, a volunteer in the UK sat down at a UK Biobank centre and over a couple of hours answered a list of questions about how they lived and then gave a blood sample. In the years since, that volunteer’s record has been kept and added to; their genome sequenced, disease biomarkers tallied, imaging data generated, all alongside every diagnosis, prescription and hospital stay, year after year, alongside half a million other people. The UK Biobank has built a unique record of health, yet one thing that’s never been included has been a detailed picture of their immune system, until now.

Tackling the scale problem

Immunology is brimming with questions, and one of the things that has perhaps held it back over the years is the lack of tools to study it in enough depth, consistency and scale to answer those complex questions properly. IMU Bioscience’s HD Immunology was built to predict what the immune system is about do, and it already addresses the first two problems in that sentence. For depth, we resolve up to 2,500 immune cell subsets and more than 100 million data points, per patient, from a single 2 ml blood sample. For consistency, our process is automated and standardised so a profile generated today can be compared against one generated next year.

That leaves scale, the third problem. And scale is essential because for any immune profile, however detailed it may be, to have clinically relevant predictive power, it needs context. Why does one patient respond to a therapy while another sees little benefit? Is a shift in one subset a warning or is it ordinary for someone of that age and sex? You can only answer those questions once you have the scale to see what normal variation looks like across a large number of people. Because only then does a departure from ‘normal’ become a signal worth following. 

Why UK Biobank

We could have built scale slowly, one cohort at a time, adding data incrementally, but partnering with UK Biobank lets fulfil our ambition of scaling more quickly and with so much additional context attached. From this September, we began profiling blood sample from an initial 80,000 UK Biobank participants, with 40,000 fresh samples and another 40,000 drawn from UK Biobank’s cryopreserved samples. Our longer-term goal is to extend profiling across the entirety of UK Biobank’s 500,000 participants.

UK Biobank stands out as the world's largest and most comprehensive biomedical database because of the additional data that sits around each sample: whole genome and exome sequencing; imaging data, from MRIs of the brain and heart to bone density scans and blood vessel ultrasounds; lifestyle and environmental data; and linked health records that now stretch across nearly two decades of follow-up. Each immune profile we generate fits into a life that has already been carefully and respectfully documented, often for years before a sample was ever run. 

The stored samples carry particular weight since some of them are the last remaining vials from those original visits, and once they’re used, they’re gone. We don’t take it lightly that UK Biobank has trusted us with them or that half a million volunteers gave their time and blood so this kind of research could happen – it’s a privilege, and a responsibility to make the most of every single one. 

From patterns to prediction

From the very start, we’ve treated the immune system as a whole system; thousands of interacting cell populations all measured together, rather than a handful of markers read in isolation. After all, you can’t find a signal in a part of the immune system you didn’t measure! This UK Biobank partnership extends that logic. We know the immune system doesn’t operate on its own, so it’s worth asking how immune states line up with genetics, with what the heart and brain look like, with how much people smoke, sleep and move, and with what happens to their health over the years that follow. 

In UK Biobank’s case, those stored samples add something fresh ones can’t. For a sample taken years ago, we can ask whether particular immune states appear earlier in the records of people who went on to develop a disease. That work begins as building associations, but that’s how a predictive signal first gets identified and that’s the very thing we plan to build on, progressively building a baseline of patterns to eventually turn into predictions. 

There’s a discovery side to this, too. When we profiled 1,015 healthy volunteers in some of our earlier work, our data showed that people have an immune age that doesn’t always match their chronological age, and that smokers’ immune systems tended to look older than their years. At UK Biobank’s scale, we can look for patterns like that across age, sex and medical history, in areas like ageing, infection, autoimmunity, immune-oncology, metabolic disease and neurodegeneration. Large population studies have rarely had the immune resolution to connect those dots and, in all honesty, we don’t know what they’ll turn up, which is a big part of why we want to look. 

Data we want AI to learn from

The first 10,000 of those UK Biobank samples are being funded through a UK Sovereign AI grant. Alongside our own profiling, those samples will also generate single-cell transcriptomic data to give a second, deeper layer of immune information from the same donors. This grant reflects a commonly held notion that biological models are limited more by the data they’re trained on than the algorithms used. Datasets stitched together from different labs, under different protocols and on different instruments can carry batch differences models learn as if they were real biology. Add to that light donor metadata and a model may be left with not quite enough information to connect a cell’s state to a person’s health. 

But we’re building this dataset the other way round, using one platform, one protocol, one cohort, with every profile linked to the genetic, imaging, lifestyle and clinical data UK Biobank already holds. It’s generated in the UK, from UK volunteers, by a UK company and governed by UK Biobank. That makes it a sovereign asset in a very practical sense, and the value it creates stays anchored in the UK research community that made it possible. 

It’s worth adding in here what AI can and can’t do. The models are incredibly sophisticated, but, as Adam Laing, our co-founder and Chief Scientific Officer put it, “sophisticated modelling becomes powerful when built on deep, consistent and biologically meaningful data. Our view is therefore data-led and the quality of the immune data comes first, since the models we rely on are only as good as what they learn from. 

Giving it back

The vast amount of data we’re set to generate doesn’t stay with us. In time, that dataset returns to UK Biobank to become an entire new immune layer available to any approved researcher through UK Biobank’s standard process – researchers won’t have to work with us to use it. 

There are far more questions about the immune system than any single company can ask, and that’s why we want to give the data back to the research community. An academic group studying long-term infection outcomes or a Pharma team validating a drug target or someone building the next immunological models – these people will have a host of questions we never thought of, and we want them to have answers. 

Our ambitions – to eventually build the most detailed and accurate database of the human immune system at the highest ever per-patient resolution, all at a population-scale – are grand, but that’s only because we know how far the benefits of achieving them could reach. 

So, coming back to those volunteers and the story their blood samples tell. For nearly two decades, their story has been missing its immune chapter, but soon, it won’t be, and neither will tens of thousands of other stories. We have our own questions for that data, but we’re just as curious about the ones other people are going to ask.