Heavy chain sequence-based classifier for the specificity of human antibodies

Area of Science:

• Immunology and Bioinformatics
• Computational Biology
• Machine Learning in Life Sciences

Background:

• Antibodies are crucial for immune response, research, and diagnostics.
• High-throughput sequencing generates vast antibody sequence data requiring analysis.
• Analyzing antibody sequences can reveal insights into antigen specificity.

Purpose of the Study:

• To develop a machine learning model for classifying antibody antigen specificity.
• To identify key features within antibody sequences that determine specificity.
• To explore the potential of computational methods for antibody discovery.

Main Methods:

• Downloaded antibody sequences from IMGT/LIGM-DB and Sequence Read Archive.
• Engineered numerical inputs from antibody heavy chain features.
• Utilized an ensemble machine learning classifier for antigen specificity prediction.
• Validated the classifier using cross-validation and a dedicated testing dataset.

Main Results:

• Achieved a macro-average AUC of 0.9246 (cross-validation) and 0.9264 (testing dataset).
• Complementarity-determining regions contributed 53.1% and framework regions 46.9% to specificity prediction.
• Amino acid mutation rates were identified as top contributing features.

Conclusions:

• The developed ensemble classifier accurately predicts antibody antigen specificity.
• Complementarity-determining regions and mutation rates are key determinants of specificity.
• This computational approach can advance mechanistic studies and the development of therapeutic antibodies.