A somatic hypermutation-based machine learning model stratifies individuals with Crohn's disease and controls.

Area of Science:

• Immunology
• Gastroenterology
• Computational Biology

Background:

• Crohn's disease (CD) is a chronic inflammatory disorder affecting the gastrointestinal tract with known immune dysregulation.
• While T cell receptor studies in CD are common, the B cell receptor (BCR) repertoire remains less understood.
• Investigating the BCR repertoire may uncover novel insights into CD pathogenesis and diagnostics.

Purpose of the Study:

• To analyze the B cell receptor (BCR) repertoire in pediatric Crohn's disease patients.
• To identify specific somatic hypermutation (SHM) patterns in the BCR repertoire associated with CD.
• To develop a machine learning (ML) model for classifying CD based on BCR features.

Main Methods:

• Collected BCR repertoire sequencing data from ileal biopsies of pediatric CD patients and controls.
• Employed a machine learning (ML) algorithm trained on BCR sequences to identify CD-specific SHM patterns.
• Applied ML classification to blood sample data from adult CD patients and controls, analyzing V gene usage, clusters, and mutation frequencies.

Main Results:

• Identified novel, CD-specific somatic hypermutation (SHM) patterns within the BCR repertoire.
• Achieved high accuracy (F1 score > 90%) in classifying CD from blood samples using ML based on BCR features like V gene usage and mutation frequency.
• Demonstrated the potential of ML algorithms to classify CD using unique BCR repertoire characteristics.

Conclusions:

• Machine learning algorithms can accurately classify Crohn's disease based on distinct B cell receptor (BCR) repertoire features.
• BCR repertoire analysis, particularly SHM patterns and V gene usage, holds significant diagnostic potential for CD.
• This research highlights the utility of ML in dissecting complex immune profiles for disease classification.