Disease diagnostics using machine learning of B cell and T cell receptor sequences.

Summary

Using receptor repertoires from 593 individuals, the authors developed an interpretable ML framework (MAchine Learning for Immunological Diagnosis) that detects infections, autoimmune diseases, vaccine responses, and severity differences from BCR/TCR sequences. Model features recapitulated known immune responses to SARS-CoV-2, illustrating translational potential for respiratory and systemic diseases.

Key Findings

• Developed an interpretable ML framework using BCR/TCR sequences from 593 individuals to screen multiple diseases or test a specific condition.
• Accurately detected specific infections, autoimmune disorders, vaccine responses, and disease severity differences.
• Model features recapitulated known immune responses to SARS-CoV-2, supporting relevance to respiratory infections.

Clinical Implications

Potential for minimally invasive blood-based diagnostics to augment current testing for respiratory pathogens, stratify disease severity, and monitor vaccine responses.

Limitations

• Prospective clinical validation and real-world workflow integration are needed
• Cohort size and disease breadth, while broad, may require expansion for rare conditions

Future Directions

Prospective trials in clinical settings, expansion across pathogens and autoimmune spectra, and head-to-head comparisons with standard diagnostics.