Its immune system houses the information of a life about the threats found: a biological role of bad. Often, perpetrators are viruses and bacteria that it has conquered; Others are undercover agents such as vaccines administered to trigger protective immune responses or even red -shaped red sandwlands trapped in immune crossfire.
Now Stanford Medicine researchers have devised a way to extract this rich internal database to diagnose diseases as diverse as COVID-19 diabetes responses to influenza vaccines. Although they imagine the approach as a way of detecting multiple diseases simultaneously, the technique based on automatic learning can also be optimized to detect complex autoims diseases and difficult to diagnose such as lupus.
In a study of almost 600 people, some healthy, others with infections that include COVID-19 or autoimmune diseases, including lupus and type 1 diabetes, the algorithm developed by researchers, called Mal-ID for automatic learning for immune diagnosis, He had remarkably successful success by identifying who had what was based only on his sequence and structures of the B and T cell receiver.
“The diagnostic tools kits that we use today do not use the internal record of the immune system of the diseases it has found,” said postdoctoral scholar Maxim Zaslavsky, PHD. “But our immune system is constantly surveys our bodies with B and T cells, which act as molecular threat sensors. Combining information from the two main arms of the immune system gives us a more complete image of the response of the immune system to the disease and the ways to autoimmunity and vaccine response. “
Zaslavsky and Erin Craig are the main authors of the study published on February 20 in Science. Professor of Pathology Scott Boyd, MD, PHD, and Associate Professor of Genetics and Informatics Anshul Kundaje, PHD, are the main authors of the research.
In addition to helping the diagnosis of difficult diseases, Mal-ID could track responses to cancer immunotherapies and subcategorize disease states so that they could help guide clinical decision making, researchers believe.
“Several of the conditions we were seeing could be significantly different at the biological or molecular level, but we describe them with broad terms that do not necessarily explain the specialized response of the immune system,” said Boyd, who co -directed the Sean N. Parker Center for Allergy and Asthma Research. “Mal-ID could help us identify subcategories of particular conditions that could give us clues about what type of treatment would be more useful for someone’s state of someone’s disease.”
Decipher the language of proteins
In a points monitoring approach, scientists used automatic learning techniques based on large language models that underlie chatgpt for the home in the receivers that recognize threats in immune cells called T cells and the commercial extremes of The antibodies (also called receptors) made by another type of immune cell called B cells. These language models look for patterns in large data sets such as book texts and Websites With enough training, you can use these patterns to predict the following word in a prayer, among other tasks.
In the case of this study, scientists applied a large protein language model, fed the millions of millions of sequences of B and T cell receptors, and used it to group receptors that share key characteristics, as determined by what is determined by The model, the model – – that could suggest similar binding preferences. Doing so could give an idea of what triggers caused a person’s immune system to mobilize, producing an army of T cells, B cells and other immune cells equipped to attack real and perceived threats.
“The sequences of these immune receptors are very variable,” said Zaslavsky. “This variability helps the immune system to detect virtually anything, but it also makes us harder Variable information with some new automatic learning techniques.
B cells and T cells represent two arms separated from the immune system, but the way they make proteins that recognize infectious agents or cells that must be removed is similar. In summary, specific DNA segments in cell genomes are randomly mix Generate unique antibody antibodies (in the case of B cells) or cell surface receptors (in the case of T cells).
The randomness of this process means that these antibodies or T -cell receptors are not adapted to recognize any specific molecule on the surface of the invaders. But its vertiginous diversity ensures that at least some join almost any foreign structure. (Autoimmunity, or an immune system attack in body tissues, is typically, but not always, avoided by a T -conditioning process and B cells pass early in the development that eliminates problematic cells).
The act of the union stimulates the cell to make many more of itself to mount a large -scale attack; The subsequent increase in the prevalence of cells with receptors that coincide with similar three -dimensional structures provides a biological fingerprint of what diseases or conditions that the immune system has been attacking.
To test their theory, the researchers gathered a data set of more than 16 million sequences of B cell receptors and more than 25 million sequences of 593 people T cell receptors with one of the six different immune states: healthy controls , people infected with SARS-COV-2 (the virus that causes COVID-19) or with HIV, people who had recently received an influenza vaccine and people with people with lupus or type 1 diabetes (both autoimmune diseases). Zaslavsky and his colleagues used their automatic learning approach to find common points among people with the same condition.
“We compare the frequencies of the use of the segment, the amino acid sequences of the resulting proteins and the way in which the model represented the” language “of the receptors, among other features,” Boyd said.
T and B cells together
The researchers found that the sequences of the T cell receiver provided the most relevant information about lupus and type 1 diabetes, while the sequences of the B cell receptor were more informative to identify HIV infection or SARS-COV-2 or The recent vaccination against influenza. In all cases, however, the combination of the results of T and B cells increased the algorithm capacity to precisely classify people by their disease, regardless of sex, age or race.
“Traditional approaches sometimes struggle to find groups of receptors that look different but recognize the same objectives,” said Zaslavsky. “But this is where large language models stand out. They can learn the grammar and specific clues of the context of the immune system as they have dominated the English grammar and context. In this way, the mal-UD can generate an understanding Internal of these sequences that give us ideas that we have not had before. “
Although the researchers developed Mal-ID in just six immunological states, they imagine that the algorithm could adapt rapidly to identify specific immunological signatures for many other diseases and conditions. They are particularly interested in autoimmune diseases such as lupus, which can be difficult to diagnose and treat effectively.
“Patients can fight for years before obtaining a diagnosis, and even then, the names that we give these diseases are as umbrella terms that overlook biological diversity behind complex diseases,” said Zaslavsky. “If we can use Mal-ID to unravel heterogeneity behind lupus or rheumatoid arthritis, that would be very clinically shocking.”
Mal-ID can also help researchers identify new therapeutic objectives for many conditions.
“The beauty of this approach is that it works even if at first we do not know completely to which molecules or structures is pointing to the immune system,” Boyd said. “We can still obtain the information simply seeing similar patterns in the way people respond. And, when deepening these answers, we can discover new addresses for research and therapies.”
Researchers at the Institute of Tropical and Public Health Switzer Sinai, Duke, the University, the University of Duke, the Swedish Medical Center, the University of Washington, the Institute of Biology of Systems, the Harvard Public School of Public Health Thard, Beth Israel Diacononess Medical Center, New York University and the Lupus Foundation of America contributed to work.
The study was funded by the National Health Institutes (Subsidies R01AI130398, R01AI127877, U19AI057229, U54CA260518, U19AI167903, 5R01 EB00198-16, UM-1 AI100645, UM1 AI144371, AI 101093 AI-086037, AI-48693, R01AI153133, R01AI137272, 3U19AI057229-17W1 COVID SUPP2, AR07375, UM1AI144292, NIDDK P30DK116074, U54CA260518, U19AI167903 R0DK1074, U54CA260518, U19AI167903, AI175771-01, R01 CA264090-01, U19 AI057229 and 1U54CA26051), the National Science Foundation, the Burroughs Wellcomme Fund, the Sunshine Foundation Gustav Floren Trust, a philanthropic gift of Eva Grove and a philanthropic gift of an anonymous donor of an anonymous.