Blood can tell you which disease is coming next - long before symptoms appear

It's not about a "magic" test that will tell a person their future diagnosis in advance. Scientists worked with statistical models of risk: they tested whether the molecular composition of blood can better predict who is more likely to develop common diseases later in life.

The study used data from 23,776 UK Biobank participants. They measured 159 metabolites - small molecules associated with metabolism - and 2,923 proteins in their blood. The researchers then compared how much such data improved the prognosis of 17 diseases compared to conventional clinical measures. The work is published in Nature Communications.

Details

Doctors typically assess disease risk by age, gender, family history, lifestyle and standard tests. These data are important, but they don't always show what's already happening inside the body at the molecular level.

The authors of the new paper decided to test whether more detailed blood data would help. To do this, they used an approach called multomics. In simple words, it is the simultaneous study of different "layers" of biology: for example, proteins and small molecules that circulate in the blood.

Proteins are particularly interesting because they are directly involved in cell function, inflammation, immune responses, metabolism and the development of many diseases. Metabolites reveal metabolic status - for example, lipid profiles and other chemical processes.

The researchers built predictive models and compared several variants. Some models took into account only familiar clinical data. Others added metabolites, proteins, or both types of molecular data at once.

The results were notable: adding "omics" improved prognosis for all 17 diseases studied. But it was proteins that were the main source of benefit. Protein-only models generally performed better than metabolite-only models for 16 of the 17 diseases. And combining proteins and metabolites produced only a small benefit compared to proteins alone.

The scientists also found molecular markers that are already well known to doctors. For example, the KLK3 protein, better known as PSA, is linked to prostate cancer. But there were also less obvious candidates - such as PRG3, which in the model appeared to be linked to skin cancer risk. Such findings may suggest new directions for future research, but they don't yet become clinical tests on their own.

Importantly, the study doesn't say, "you can tell exactly what a particular person will get sick from blood." What it does show is that when a detailed molecular profile of the blood is added to routine data, risk prediction in a large group of people becomes more accurate.

Why it matters

If such models are confirmed in other studies, they could help medicine move toward earlier and more personalised disease prediction. Ideally, a doctor will not only be able to see a person's current vitals, but also notice molecular risk signals long before overt symptoms.

This is especially important for diseases that take a long time to develop invisibly. The earlier it is possible to realise that a person is at high risk, the greater the opportunities for surveillance, prevention and timely treatment.

But there is still a long way to go before it can be applied in a conventional clinic. Such models need to be tested in different populations, in different countries and in real medical settings. UK Biobank is a very valuable resource, but its members do not perfectly reflect the whole population. In addition, the researchers only analysed data collected at the start, not continuous changes in blood over time.

So the main conclusion is a cautious one: blood may indeed contain early molecular clues about future disease risk, but turning this into a ready-made "test for all diseases" is not yet possible.

Background

The idea of early disease prediction from blood molecules has been actively developed in recent years. Scientists are increasingly studying not just one indicator, but large data sets: proteins, metabolites, genetic variants, gene activity and other biological layers.

This approach is necessary because most common diseases do not arise from a single cause. Heredity, age, metabolism, inflammation, medications, lifestyle, social conditions, and a host of other factors influence risk.

The new work is important in scale. The authors say it is one of the largest studies to simultaneously evaluate the contribution of metabolomics and proteomics to predicting future disease. The paper was published on 9 May 2026; the Nature Communications page indicates that an early unedited version of the manuscript is available before final editing.

Source

Jiawen Du, Muqing Zhou, Hanling Wang et al, "Multi-omics integration predicts the incidence of 17 diseases in the UK Biobank", Nature Communications, 2026.

The study used data from 23,776 UK Biobank participants with baseline values of 159 NMR metabolites and 2,923 Olink proteins in plasma. The authors compared prediction models for 17 diseases and showed that the addition of molecular data improved risk prediction over clinical indicators alone. Protein profiles were found to be most informative in most cases.