Statistical classifiers for diagnosing disease from immune repertoires: a case study using multiple sclerosis.

Area of Science:

• Immunology
• Bioinformatics
• Computational Biology

Background:

• Deep sequencing of lymphocyte receptor repertoires enables comprehensive profiling of immune cell populations.
• Identifying repertoire sequence patterns associated with clinical phenotypes offers novel diagnostic and prognostic approaches for immune-driven diseases.
• Prior methods focused on repertoire-level statistics, overlooking information within individual immune receptors.

Purpose of the Study:

• To develop a novel method utilizing machine learning to analyze individual immune receptor sequences for disease diagnosis.
• To apply this method to diagnose relapsing remitting multiple sclerosis (RRMS), an autoimmune disease known for diagnostic challenges.

Main Methods:

• Developed a novel method incorporating innovative approaches for immune receptor sequence diversity and machine learning.
• Utilized biochemical features of B cell receptor heavy chain complementarity determining region 3 as input.
• Employed maximum likelihood optimization for fitting a statistical classifier.

Main Results:

• The statistical classifier achieved 87% accuracy in diagnosing RRMS on training data (N=23) using leave-one-out cross-validation.
• The classifier demonstrated 72% accuracy on independent test data from a separate study (N=102).
• Identified a potential diagnostic biochemical motif in RRMS patient antibodies.

Conclusions:

• This study presents the first statistical learning approach applied to immune repertoires for disease diagnosis.
• The developed repertoire-based classifier for RRMS diagnosis exhibits high accuracy, comparable to clinical experts.
• The findings suggest potential insights into the disease mechanisms of RRMS through identified antibody motifs.