Evaluating predictive patterns of antigen-specific B cells by single-cell transcriptome and antibody repertoire...

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• Traditional antibody discovery relies on extensive experimental screening of B cells.
• Machine learning (ML) can accelerate this process by predicting antigen-specific B cells, but requires labeled training data.

Purpose of the Study:

• To develop and evaluate ML models for predicting antigen-specific B cells using gene expression and antibody repertoire data.
• To identify key features and model types that best predict antigen specificity.

Main Methods:

• Generated a dataset of single-cell transcriptome and antibody repertoire sequencing from immunized mouse B cells.
• Labeled B cells as antigen-specific or non-specific via experimental selection.
• Performed differential gene expression analysis to find specificity-associated patterns.
• Benchmarked various ML models (linear and non-linear) using gene expression and antibody sequence features.
• Assessed transfer learning with protein language models (PLMs).

Main Results:

• Identified distinct gene expression patterns correlated with antigen specificity.
• Gene expression-based ML models demonstrated superior performance in predicting antigen specificity compared to antibody sequence-based models.
• Transfer learning approaches showed potential but did not surpass gene expression features.

Conclusions:

• Gene expression profiles are powerful predictors of B cell antigen specificity.
• ML models leveraging gene expression data offer a promising computational approach to accelerate antibody discovery.
• This work provides a valuable dataset and benchmarks for future ML-driven antibody discovery efforts.