MaxSnippetModel

MaxSnippetModel analyzes immune-repertoire deep-sequencing data to detect biochemical features in CDR3 regions of B cell receptor heavy chains and classify patients for diagnosis of relapsing-remitting multiple sclerosis (RRMS).

Key Features:

• Comprehensive Profiling: Uses deep sequencing to profile lymphocyte receptor repertoires, capturing millions of individual immune receptor sequences.
• CDR3 Biochemical Analysis: Focuses on biochemical features encoded in the complementarity determining region 3 (CDR3) of B cell receptor heavy chains rather than repertoire-level summary statistics.
• Machine Learning Integration: Integrates machine learning approaches with maximum likelihood optimization to fit a detector function and construct classifiers.
• Diagnostic Performance: Reported 87% accuracy in leave-one-out cross-validation on training data (N = 23) and 72% accuracy on independent validation data from a separate study (N = 102).
• First-of-its-Kind Application: Applies statistical learning to immune repertoires specifically for diagnosis of relapsing-remitting multiple sclerosis (RRMS).

Scientific Applications:

• RRMS Diagnosis: Distinguishes relapsing-remitting multiple sclerosis (RRMS) from other neurological diseases using repertoire-based statistical classification.
• Diagnostic Motif Identification: Identifies a diagnostic biochemical motif in antibodies of RRMS patients by analyzing mutation frequencies and sequence diversity within VH4-containing genes in B cells.
• Clonal Composition Analysis: Characterizes clonal composition and sequence patterns in lymphocyte repertoires to support diagnosis and investigation of immune-mediated disease processes.

Methodology:

Input data comprise biochemical features from CDR3 regions of B cell receptors derived from deep sequencing. A machine learning–based detector function is fit using maximum likelihood optimization to generate a repertoire-based statistical classifier. Performance was evaluated with leave-one-out cross-validation (N = 23) and independent validation (N = 102).