Language Model Embedding Classifiers Enable Identification of Multiple Sclerosis-Associated BCRs and Repertoires

Area of Science:

• Immunology
• Computational Biology
• Neurology

Background:

• Multiple sclerosis (MS) is a chronic inflammatory demyelinating disease affecting over 2 million people globally.
• MS diagnosis, categorization, and treatment remain challenging due to its complex autoimmune component involving B-cell receptors (BCRs) and immunoglobulins (Igs).

Purpose of the Study:

• To develop and validate advanced computational models for identifying B-cell receptor (BCR) sequences associated with multiple sclerosis (MS).
• To establish a methodological foundation for BCR-based MS detection and facilitate the study of disease-associated antibodies.

Main Methods:

• Reanalysis of publicly available RNA sequencing data from MS patients to extract over 11 million BCR immunoglobulin heavy chain (IGH) sequences.
• Development of a novel decoder-only BCR DNA embedding model demonstrating superior performance on sequence embedding tasks.
• Training a language model classifier to identify MS-associated BCR sequences and distinguishing MS repertoires from healthy or other disease controls.

Main Results:

• The developed BCR DNA embedding model outperforms existing state-of-the-art models.
• The language model classifier successfully differentiates MS patient B-cell receptor repertoires from those of healthy individuals and patients with other diseases.
• The models accurately rank known MS-associated myelin-binding IgG sequences.

Conclusions:

• The study presents a robust computational framework for BCR-based detection of multiple sclerosis.
• These findings pave the way for improved diagnostic tools and a deeper understanding of MS pathophysiology through antibody analysis.