Predicting antibody binders and generating synthetic antibodies using deep learning

Area of Science:

• Biotechnology
• Computational Biology
• Immunology

Background:

• The antibody drug field traditionally relies on laboratory methods for candidate discovery and engineering.
• Informatics and machine learning, particularly deep learning, are emerging as powerful tools in biomedical research.
• Advances in microfluidics and next-generation sequencing have generated large antibody repertoire datasets, enabling deep learning applications.

Purpose of the Study:

• To explore the application of deep learning for analyzing antibody sequences and improving antibody discovery and engineering.
• To develop and train convolutional neural network (CNN) models for classifying antibody binders and non-binders against cancer immunotherapy targets.
• To generate synthetic antibodies using generative deep learning models.

Main Methods:

• Utilized microfluidics, yeast display, and deep sequencing to generate antibody sequences targeting PD-1 and CTLA-4.
• Encoded antibody complementarity-determining regions (CDRs) into images for CNN model training.
• Performed in silico mutagenesis for model interpretability and identified key CDR3 residues.
• Employed generative adversarial network (GAN) models to create synthetic antibody sequences.

Main Results:

• Developed CNN models capable of classifying antibody binders and non-binders with high accuracy.
• Identified critical CDR3 residues influencing antibody binding through in silico mutagenesis.
• Generated novel, synthetic antibody sequences against PD-1 and CTLA-4 using GANs, exhibiting realistic CDR3 characteristics.
• Demonstrated the potential of deep learning to uncover patterns in antibody sequences.

Conclusions:

• Deep learning methods are effective for mining and learning patterns within antibody sequence data.
• This approach offers significant insights for antibody engineering, optimization, and the discovery of novel therapeutic antibodies.
• The study highlights the integration of computational methods with experimental data for advancing antibody-based drug development.