Revolutionary Blood Test Detects Early-Stage Solid Tumors with High Accuracy Using AI

New blood test detects early-stage solid tumors with high accuracy

Spanish National Cancer Research Centre (CNIO)Jul 31 2025 Current methods for cancer diagnosis are based on identifying biomarkers - molecules that reveal a particular state or process in the body - produced by the tumour or associated proteins. Not surprisingly, these markers are more abundant once the tumour has already developed significantly. And the more advanced the tumour, the more difficult it is to find effective treatment options. Now, a team led by Gonçalo Bernardes, head of the Translational Chemical Biology Group at the Spanish National Cancer Research Centre (CNIO), has developed a test that can detect early-stage solid tumours with just a blood sample. In addition, the test also provides information relevant to the choice of treatment. The study has been published in the scientific journal Nature Communications. Studying the proteins that react to cancer To achieve this early detection, the team led by the Portuguese researcher focused the test not on the markers produced by the tumour, but on the body's defensive reaction to the cancer. Since the 19th century it has been known that the emergence of cancer cells causes changes in the immune system, and it was also known that these changes are more intense in cancer's earliest stages. But they had never been used for diagnosis. The new study focuses on them, specifically on the changes in blood proteins derived from cancer's disruption of the immune system. 'Our approach,' explains Gonçalo Bernardes, 'has proven particularly effective in detecting tumors at an early stage, which is crucial because, if we detect them early, we can treat many types of cancer'. Artificial intelligence to search for patterns But this approach posed a problem to the team: human blood contains more than 5,000 proteins, which makes it extremely difficult to analyse. So they used bioinformatics analysis and narrowed the scope of the study to five amino acids: lysine, tryptophan, tyrosine, cysteine and cysteine not bound to disulphide bonds. They then subjected the sample to reactions that emit fluorescence when light is applied to them - fluorogenic reactions - and revealed the exact concentration of each of these amino acids in the plasma. Using the artificial intelligence tool machine learning, they identified patterns in these concentrations that could be translated into diagnostic signals. As they explain in the published article, they applied this technique to samples from 170 patients and were able to identify 78% of cancers with a 0% false positive rate. Bernardes also points out that the test is easy to use, requiring only a small blood sample and the use of simple reagents that can be found in any hospital. To make the diagnosis, teh team led by Bernardes, also a professor at the University of Cambridge (UK), is developing a platform that will analyse the data. Other diseases and treatment response The samples studied so far did not belong exclusively to people with cancer: 'It is very important to note,' says Gonçalo Bernardes, "that by analysing samples from patients with other diseases, we have found that the signals are different. For example, the immune signals of a person with SARS-Covid are different from the signals of a person with cancer, as are the signals of different types of cancer and even cancer in its different stages. We can identify all of that with our test. And these unique signals for each type of cancer also provide other information of enormous interest for clinical practice: whether or not the patient will respond to certain treatments. The article describes that the test predicted with 100% accuracy that a patient would not respond to anti-metastatic treatment. When it predicted that a patient would respond, the accuracy was 87%. Therefore, the authors claim that the test could also be used for precision medicine in the choice of treatments. A sample of 170 patients has been sufficient to get the study this far, but the researcher acknowledges that much more data is needed to complete the commercial development of the test. To this end, two clinical trials are already underway in the UK - funded by the UK's national health system - and a number of other trials are underway in countries such as the US and China. Once developed, the platform is expected to be commercialised through a spin-off company in Cambridge called Proteotype Ltd, which Bernardes co-founded with other authors. Spanish National Cancer Research Centre (CNIO)

Blood test analyzes immune protein changes to detect early-stage tumors and guide treatment

Current methods for cancer diagnosis are based on identifying biomarkers -- molecules that reveal a particular state or process in the body -- produced by the tumor or associated proteins. Not surprisingly, these markers are more abundant once the tumor has already developed significantly. And the more advanced the tumor, the more difficult it is to find effective treatment options. Now, a team led by Gonçalo Bernardes, head of the Translational Chemical Biology Group at the Spanish National Cancer Research Centre (CNIO), has developed a test that can detect early-stage solid tumors with just a blood sample. In addition, the test also provides information relevant to the choice of treatment. The study has been published in the journal Nature Communications. Studying the proteins that react to cancer To achieve this early detection, the team led by the Portuguese researcher focused the test not on the markers produced by the tumor, but on the body's defensive reaction to the cancer. Since the 19th century, it has been known that the emergence of cancer cells causes changes in the immune system. It was also known that these changes are more intense in cancer's earliest stages, but they have never been used for diagnosis. The new study focuses on them, specifically on the changes in blood proteins derived from cancer's disruption of the immune system. "Our approach," explains Bernardes, "has proven particularly effective in detecting tumors at an early stage, which is crucial because, if we detect them early, we can treat many types of cancer." Artificial intelligence to search for patterns But this approach posed a problem to the team: human blood contains more than 5,000 proteins, which makes it extremely difficult to analyze. So they used bioinformatics analysis and narrowed the scope of the study to five amino acids: lysine, tryptophan, tyrosine, cysteine and cysteine not bound to disulfide bonds. They then subjected the sample to reactions that emit fluorescence when light is applied to them -- fluorogenic reactions -- and revealed the exact concentration of each of these amino acids in the plasma. Using the artificial intelligence tool machine learning, they identified patterns in these concentrations that could be translated into diagnostic signals. As they explain in the published article, they applied this technique to samples from 170 patients and were able to identify 78% of cancers with a 0% false positive rate. Bernardes also points out that the test is easy to use, requiring only a small blood sample and the use of simple reagents that can be found in any hospital. To make the diagnosis, the team led by Bernardes, also a professor at the University of Cambridge (UK), is developing a platform that will analyze the data. Other diseases and treatment response The samples studied so far did not belong exclusively to people with cancer: "It is very important to note," says Bernardes, "that by analyzing samples from patients with other diseases, we have found that the signals are different. For example, the immune signals of a person with SARS-COVID are different from the signals of a person with cancer, as are the signals of different types of cancer and even cancer in its different stages. We can identify all of that with our test." And these unique signals for each type of cancer also provide other information of enormous interest for clinical practice: whether or not the patient will respond to certain treatments. The article describes that the test predicted with 100% accuracy that a patient would not respond to anti-metastatic treatment. When it predicted that a patient would respond, the accuracy was 87%. Therefore, the authors claim that the test could also be used for precision medicine in the choice of treatments. A sample of 170 patients has been sufficient to get the study this far, but the researcher acknowledges that much more data is needed to complete the commercial development of the test. To this end, two clinical trials are already underway in the UK.