Disease diagnostics using machine learning of immune receptors

Area of Science:

• Immunology
• Bioinformatics
• Machine Learning

Background:

• Current clinical diagnosis relies on physical exams, history, lab tests, and imaging, often overlooking the immune system's record of antigen exposures.
• The B cell and T cell receptor repertoire encodes a history of immune encounters, offering a potential diagnostic resource.

Approach:

• Developed MAchine Learning for Immunological Diagnosis (Mal-ID), an interpretive framework analyzing immune receptor datasets from 593 individuals.
• Utilized machine learning to screen for multiple illnesses simultaneously or pinpoint specific conditions.

Key Points:

• Mal-ID successfully detects specific infections (e.g., SARS-CoV-2, Influenza, HIV), autoimmune disorders (Systemic Lupus Erythematosus, Type-1 Diabetes), and vaccine responses.
• The model's human-interpretable features align with known immune responses and highlight antigen-specific receptors.
• Demonstrated distinct autoreactivity patterns in Systemic Lupus Erythematosus and Type-1 Diabetes.

Conclusions:

• The Mal-ID framework offers a novel approach to clinical diagnosis by leveraging immunological data.
• This analysis framework has broad potential for scientific and clinical interpretation of human immune responses.
• Mal-ID can screen for multiple conditions or precisely diagnose single diseases, enhancing diagnostic capabilities.