Computational models for the classification of antibody specificity using heavy chain features

Area of Science:

• Immunoinformatics
• Computational Biology
• Machine Learning

Background:

• Antibody specificity is determined by heavy chain sequences, crucial for immune defense, research, and diagnostics.
• High-throughput sequencing generates vast antibody data requiring advanced analytical methods.

Purpose of the Study:

• To develop computational models for classifying antibodies into predefined antigen classes based on heavy chain sequences.
• To identify sequence-based features driving antibody-antigen specificity.

Main Methods:

• Extracted physicochemical, structural, sequence order, and evolutionary features from antibody heavy chain sequences.
• Utilized machine learning (CatBoost, stacking) and deep learning (Feature-Based Transformer) models for classification.
• Applied SHAP framework to determine feature importance for antigen specificity prediction.

Main Results:

• Achieved highest accuracy of 0.7803 with a stacking model and 0.7713 with CatBoost.
• Identified sequence order and evolutionary information as predominant features, with cysteine (Cys) being most influential.
• Highlighted contributions of specific amino acids and antibodies to key features.

Conclusions:

• Demonstrated the effectiveness of machine and deep learning for antibody classification based on sequence data.
• Provided sequence-based insights into antibody specificity determinants.
• Implications for understanding, isolating, and developing therapeutic antibodies.