AI-assisted diagnosis for immunological disease

(Press-News.org) A novel machine learning framework – Mal-ID – can decipher an individual’s immune system’s record of past infections and diseases, according to a new study, providing a powerful tool with the potential for diagnosing autoimmune disorders, viral infections, and vaccine responses with precision. Traditional clinical diagnostic methods for autoimmune diseases or other immunological pathologies tend to rely on a combination of physical examination, patient history, and various laboratory testing for cellular or molecular abnormalities – a lengthy process often complicated by initial misdiagnoses and ambiguous systems. These approaches make limited use of data from the patient’s individual adaptive immune system’s B cell receptors (BCRs) and T cell receptors (TCRs). In response to pathogens, vaccines, and other antigenic stimuli, BCR and TCR repertoires undergo changes through clonal expansion, somatic mutation, and selective reshaping of immune cell populations. Sequencing BCRs and TCRs could provide a comprehensive diagnostic tool, potentially enabling simultaneous detection of infectious, autoimmune, and immune-mediated diseases in a single test. However, the extent to which immune receptor repertoire sequencing alone can reliably and broadly classify diseases remains uncertain.

 

To address this, Maxim Zaslavsky and colleagues developed Mal-ID (MAchine Learning for Immunological Diagnosis) – a 3-model machine learning framework that analyzes immune receptor datasets to identify signatures of infectious and immunological diseases and vaccine responses in patients. Zaslavsky et al. trained Mal-ID on BCR and TCR data systematically collected from 593 individuals, including patients with COVID-19, HIV, and type-1 diabetes, as well as influenza vaccine recipients and healthy controls. According to the findings, Mal-ID effectively distinguished six distinct disease states in 550 paired BCR and TCR samples with a multiclass AUROC score of 0.986, indicating exceptionally high classification accuracy. This metric reflects the model’s ability to rank positive cases above negative ones across all disease comparisons. Although the model was successful in differentiating COVID-19, HIV, lupus, T1D, and healthy individuals – illustrating its potential as a powerful diagnostic tool – Zaslavsky et al. note that the approach still needs to be refined, using clinical information, before it could be used with confidence in clinical applications.

END