Scientists are learning to read thousands of proteins circulating in the blood to identify patterns associated with diseases years before diagnosis. The approach could transform early detection—but turning a promising prediction into a reliable medical test remains a major challenge.
A routine blood test can reveal a great deal about a person’s health. It can measure glucose, cholesterol, blood cells and markers of organ function. But what if a blood sample could reveal clues about a disease that has not yet produced noticeable symptoms?
That possibility is attracting growing attention from researchers studying the human proteome: the enormous collection of proteins found in the body.
In recent years, scientists have begun combining large-scale protein measurements with medical records and machine learning to investigate whether subtle changes in the blood can predict future illness. Some of the findings are striking. Research published in Nature Medicine in 2024 showed that patterns involving relatively small numbers of blood proteins could improve predictions for dozens of diseases, sometimes years before diagnosis.
The implications are significant. Instead of waiting for symptoms to appear, doctors might eventually be able to identify certain people who need closer monitoring or additional testing.
But there is an important distinction between predicting an increased risk and detecting a disease with certainty. The technology is promising, yet it still faces substantial scientific and clinical hurdles.
What is proteomics, and why does it matter?
Proteomics is the large-scale study of proteins, including their abundance, structure and biological functions.
Proteins perform much of the body’s essential work. They carry chemical messages, regulate immune responses, transport substances, repair tissues and help cells communicate. Their levels can change in response to infection, inflammation, ageing, metabolic changes and developing disease.
That makes blood an especially valuable source of information.
Although a blood sample contains thousands of different proteins, conventional laboratory tests measure only a relatively small selection. Large-scale proteomic technologies can examine thousands of proteins simultaneously, allowing researchers to search for combinations that might be associated with a particular disease.
The key is that a disease may not be reflected by a single abnormal protein. Instead, it may produce a distinctive pattern involving several proteins, each contributing a small amount of information.
Researchers can use statistical models to identify these patterns and estimate how strongly they are associated with future diagnoses.
The resulting protein signature is rather like a biological fingerprint. It does not necessarily reveal the entire process taking place inside the body, but it may provide clues that would be difficult to identify through conventional testing alone.
A major study examined thousands of proteins
One important study, published in Nature Medicine in 2024, analysed data from 41,931 participants in the UK Biobank Pharma Proteomics Project.
Researchers measured approximately 3,000 plasma proteins and investigated whether their patterns could predict the development of 218 different diseases over the following decade.
Their findings suggested that models using as few as five to 20 proteins could outperform predictions based on basic clinical information for 67 diseases. For 52 diseases, the protein-based models also performed better than models incorporating conventional clinical information and available laboratory measurements.
Conditions examined included multiple myeloma, non-Hodgkin lymphoma, motor neuron disease, pulmonary fibrosis and dilated cardiomyopathy.
The researchers also tested selected models in an independent cohort, providing additional evidence that some of the associations could be reproduced outside the original analysis.
Read the original study in Nature Medicine.
These results demonstrate why proteomics has attracted so much interest. A relatively small collection of protein measurements may contain information that is not fully captured by a patient’s age, lifestyle or routine laboratory results.
However, the study was designed to investigate prediction, not to establish a universal blood test for all these conditions. Its findings identify promising possibilities for future clinical development rather than a ready-made screening programme.
Can proteins reveal disease years before symptoms appear?
In some cases, researchers have found protein patterns associated with diagnoses many years later.
One example comes from research published in Nature Communications in December 2025. Scientists analysed plasma proteins from 39,634 UK Biobank participants to investigate whether future Crohn’s disease could be predicted before diagnosis.
Crohn’s disease is a chronic inflammatory bowel disease that can cause abdominal pain, diarrhoea, fatigue and other symptoms. Diagnosis can be delayed because early symptoms may be intermittent or resemble other digestive conditions.
The researchers identified 44 proteins associated with subsequent Crohn’s disease and developed a prediction model using nine of them. In the study, the model showed predictive value in separate testing cohorts, with signals detectable up to 16 years before diagnosis in the UK Biobank analysis.
The study also found that combining protein measurements with clinical information improved predictive performance.
Read the Crohn’s disease research in Nature Communications.
This does not mean that researchers can reliably identify every future Crohn’s disease patient 16 years in advance. Nor does it establish that screening healthy populations with this model would improve outcomes.
Instead, the findings suggest that biological changes associated with the disease may begin long before a person receives a formal diagnosis.
That distinction matters because an early warning is useful only if doctors can interpret it accurately and act on it in ways that benefit patients.
What can the proteins actually reveal?
Different diseases involve different biological processes, so there is no single protein signature that explains every condition.
Some proteins may reflect inflammation or immune-system activity. Others can indicate tissue damage, changes in metabolism or alterations in how cells communicate.
In multiple myeloma, a cancer involving plasma cells in the bone marrow, researchers identified protein predictors associated with pathways relevant to the disease. This provides a useful example of how prediction models can also generate scientific clues about the biology underlying a condition.
Yet a protein associated with a future diagnosis is not necessarily causing the disease.
It could be an early consequence of a disease process that has already begun, a marker of an underlying risk factor, or a signal shared by several conditions. Researchers need additional experiments and clinical evidence to distinguish these possibilities.
Understanding the biology behind a protein signature may help scientists develop better tests and identify potential treatment targets. But a predictive association alone cannot establish a cause-and-effect relationship.
Why false positives remain a serious problem
The prospect of detecting disease before symptoms appear sounds appealing, but screening people who feel healthy presents a difficult statistical challenge.
Suppose a test is designed to identify people at risk of a relatively rare disease. Even if it performs reasonably well, many positive results could come from people who will never develop that disease.
This happens because the proportion of positive results that represent genuine future cases depends not only on the test’s accuracy but also on how common the disease is in the population being screened.
False positives can lead to anxiety, repeat blood tests, imaging, invasive investigations and unnecessary medical costs. False negatives can provide false reassurance to people who actually need further assessment.
A model that works well in a research cohort may also perform differently in a general population. Participants in large biobanks may differ from the people who would eventually be offered screening in routine healthcare.
Age, existing medical conditions, medication use, ethnicity and other factors can influence protein levels. Models therefore need to be tested in diverse populations and in the settings where they would actually be used.
Researchers must also determine whether the predictions add meaningful information beyond existing tests. A sophisticated protein panel is not necessarily useful if a simpler, cheaper investigation provides the same information.
From research discovery to a real blood test
Before proteomic prediction becomes routine medicine, researchers must establish reliable thresholds, validate the tests independently and demonstrate that acting on the results improves patient care.
They also need laboratory methods that can measure the relevant proteins accurately and consistently across different hospitals. A promising research signature may not translate directly into a standardised clinical assay.
Another challenge is deciding what doctors should do when a test identifies elevated risk but no symptoms or other evidence of disease are present.
For some conditions, closer monitoring could allow earlier treatment. For others, there may be no proven intervention that changes the outcome when a disease is detected at such an early stage.
The ethical implications also deserve attention. Information suggesting an increased future disease risk could affect a person’s psychological wellbeing and raise questions about privacy, consent and how medical information is used.
A useful screening programme must therefore do more than predict illness. It must demonstrate that the benefits of earlier detection outweigh the harms of false alarms and unnecessary investigations.
Could this change the future of preventive medicine?
Large-scale proteomics offers a new way to investigate what is happening inside the body before illness becomes obvious.
Instead of relying exclusively on symptoms or individual laboratory measurements, doctors may eventually have access to more detailed biological risk profiles. These could help identify patients who need closer observation, guide decisions about further testing and support research into the earliest stages of disease.
The strongest evidence so far shows that protein signatures can improve risk prediction for selected conditions. It does not yet establish that one blood test can reliably predict every major disease or that widespread proteomic screening would improve health outcomes.
The next step is careful clinical validation: determining which signatures work consistently, which patients benefit and what actions should follow a positive result.
A blood sample may contain clues about a disease years before a doctor names it. The challenge now is learning which clues matter—and how to use them without mistaking a warning signal for a diagnosis.
