Machine Learning Tool Enables AI-Assisted Diagnosis of Immunological Diseases

Traditional diagnostic methods for autoimmune diseases and other immunological conditions typically combine physical examinations, patient history, and laboratory tests to detect cellular or molecular abnormalities. However, this process is often time-consuming and complicated by misdiagnoses and ambiguous symptoms. These methods generally do not take full advantage of data from the patient’s adaptive immune system, particularly from B cell receptors (BCRs) and T cell receptors (TCRs). In response to infections, vaccines, and other antigenic stimuli, BCR and TCR repertoires are altered through clonal expansion, somatic mutation, and the reshaping of immune cell populations. Sequencing these immune receptors has the potential to provide a more comprehensive diagnostic tool, enabling the detection of infectious, autoimmune, and immune-mediated diseases in one test. However, it remains uncertain how reliably and broadly immune receptor repertoire sequencing can classify diseases on its own.

A team of researchers at Stanford University (Stanford, CA, USA) has created an innovative machine learning framework called Mal-ID that can interpret an individual’s immune system record of past infections and diseases. This model provides a promising new tool for diagnosing autoimmune disorders, viral infections, and vaccine responses with precision. Mal-ID, which stands for MAchine Learning for Immunological Diagnosis, is a three-model framework that analyzes immune receptor datasets to identify patterns associated with infectious diseases, autoimmune conditions, and vaccine responses. The model was trained using BCR and TCR data collected from 593 individuals, including patients with COVID-19, HIV, type-1 diabetes, as well as individuals who received the influenza vaccine and healthy controls.

The findings, published in Science, demonstrate that Mal-ID successfully identified six distinct disease states in 550 paired BCR and TCR samples, achieving a multiclass AUROC score of 0.986, which indicates exceptionally high classification accuracy. This score reflects the model’s ability to accurately rank positive cases above negative ones across various disease comparisons. The model’s ability to distinguish between conditions such as COVID-19, HIV, lupus, type-1 diabetes, and healthy controls highlights its potential as a powerful diagnostic tool. However, the researchers noted that further refinement, incorporating clinical information, is necessary before the approach can be reliably used in clinical settings.