Based on immune cells

New AI algorithm to detect diseases

The AI tool is designed to precisely recognize diseases based on immune cells.

(Bild: fadi (computergeneriert))

AI researchers are celebrating a medical breakthrough: a new AI algorithm developed by a team of US researchers can detect infections and autoimmune diseases with great accuracy based on typical changes in B and T cells and thus recognize diseases.

Maxim Zaslavsky from the US Stanford University and his co-authors have reported this in the scientific journal "Science". The diseases are read from the gene sequence of the immune cell receptors.

The background: Currently, the diagnosis of an infectious disease is usually based on a search for the pathogen as the cause. This often means a long wait until a culture has been created and evaluated. Antibody reactions are also often delayed. In the case of autoimmune diseases such as type 1 diabetes, the disease is diagnosed via the metabolic consequences; in the case of rheumatism, with considerable uncertainty, via the combination of numerous signs of the disease, including laboratory values (rheumatoid factors).

Search for specific receptor properties
However, the scientists, including experts from the Tropical Institute at the University of Basel, pursued a new idea: Diseases trigger specific reactions in the immune system of those affected. For example, very specific receptors are formed on B and T immune cells. The experts' idea: it may be possible to read from the receptors what the body's own defense system is currently dealing with.

Our immune system constantly monitors our body with B and T cells, which act like molecular threat sensors.

Maxim Zaslavsky, Stanford University

"Our immune system constantly monitors our body with B and T cells, which act like molecular threat sensors. The combination of information from these two main areas of the immune system provides us with a more comprehensive picture of the immune system's response to diseases and the processes that lead to autoimmunity (autoimmune diseases; note) or vaccination reactions, for example," Zaslavsky was quoted as saying in a press release from the Californian university on the publication of the research work.

AI system fed with 30 million data packets
The researchers therefore sequenced certain parts of the genes for the B and T cell receptors in patients with infections or autoimmune diseases. These receptors are there to detect pathogens or, in the case of autoimmune diseases, cause misdirected defense reactions against the body's own tissue. The scientists therefore used the AI software Mal-ID ("MAchine Learning for Immunological Diagnosis") to create a program that would recognize typical changes in the receptors of immune cells for some diseases.

"In a pilot study, Mal-ID analyzed the sequence data of 16.2 million B-cell receptors and 23.5 million T-cell receptors. They came from blood samples from 593 people, 63 of whom were infected with SARS-CoV-2 and 95 with the HI virus," reported the German Medical Journal. 86 of the test subjects suffered from an autoimmune disease (lupus erythematosus), 92 from type 1 diabetes (also an autoimmune disease). 37 test subjects had been vaccinated against influenza. The control group consisted of 217 test subjects who were not affected.

Almost one hundred percent accuracy
The result: Mal-ID detected individual diseases such as SARS-CoV-2 infections, HIV infections, lupus, type 1 diabetes as well as the previous influenza vaccination with almost 100% sensitivity (detection of affected persons) and specificity (exclusion of a disease in the absence of a disease). One difference depending on the type of disease: The gene sequences of the receptors of the test subjects' B cells identified the HIV and Covid-19 infections as well as the influenza vaccination. The T-cell receptors identified those affected by lupus erythematosus and diabetes.

"As the cost of sequencing genes has fallen significantly in recent years, the method could become interesting for clinical diagnostics. This applies in particular to autoimmune diseases, which are often only diagnosed after months or years of searching," wrote the German Medical Journal.

Algorithm easily adaptable
Although the researchers have so far only developed Mal-ID on the basis of six diseases or immunological conditions (vaccination), they assume that the algorithm can be quickly adapted to identify immunological signatures that are specific to many other diseases and conditions. This is particularly true for complex autoimmune diseases, including rheumatoid arthritis and chronic polyarthritis.

"Patients often have to struggle for years before they receive a diagnosis, and even then the names we give these diseases are like generic terms that overlook the biological diversity behind complex diseases," said Zaslavsky. "If we could use Mal-ID to decipher the heterogeneity behind lupus or rheumatoid arthritis, that would have a big impact."