Learning antibody sequence constraints from allelic inclusion.

Area of Science:

• Immunology
• Computational Biology
• Genomics

Background:

• Antibody sequences exhibit high diversity but are constrained by expression and reactivity requirements.
• Identifying antibody sequences that violate these constraints is challenging.
• Allelic inclusion, where B cells express two antibody light chains, offers a novel data source.

Purpose of the Study:

• To develop a machine-learning framework using allelic inclusion data to identify abnormal antibody sequences.
• To predict antibody properties such as polyreactivity, surface expression, and mutation usage.
• To investigate heavy chain selection forces and the impact of surrogate light chain pairing in mice.

Main Methods:

• Utilized human single-cell sequencing data to identify instances of B cell allelic inclusion.
• Trained machine-learning models to recognize abnormal antibody sequences associated with allelic inclusion.
• Evaluated model performance in predicting antibody properties against existing methods.
• Analyzed mouse heavy chain sequences to assess selection forces and light chain influences.

Main Results:

• Machine-learning models trained on allelic inclusion data successfully identified abnormal antibody sequences.
• These models demonstrated superior performance in predicting polyreactivity, surface expression, and mutation usage compared to conventional methods.
• Selection forces on mouse heavy chains were investigated, revealing a significant impact of surrogate light-chain pairing on diversity.

Conclusions:

• Leveraging allelic inclusion data provides a powerful new approach for identifying problematic antibody sequences.
• Machine learning applied to this data enhances the prediction of critical antibody functional properties.
• Understanding light-chain influences on heavy-chain diversity offers insights into antibody repertoire formation.