High-throughput machine learning-aided antibody discovery for cell surface antigens.

Area of Science:

• Biotechnology
• Immunology
• Computational Biology

Background:

• Machine learning (ML) requires high-quality antibody-antigen interaction data for effective antibody design and selection.
• Existing datasets often lack the necessary curation and format for seamless ML integration.
• Antibody discovery pipelines can be significantly accelerated by leveraging advanced computational approaches.

Purpose of the Study:

• To develop a synthetic antibody library optimized for machine learning (ML) integration.
• To generate a diverse dataset of antibody sequences and their corresponding antigen interactions.
• To establish an ML-compatible framework for accelerating antibody discovery and development.

Main Methods:

• A synthetic Fab yeast display library was engineered with diverse complementary determining region heavy chain 3 (CDRH3) loop sequences.
• The library incorporated features from human B cell repertoires in an antigen recognition module (ARM) format.
• The library was screened against ten cell surface antigens, including PD-L1, TIGIT, and ROBO1, followed by ML analysis of sequencing data.

Main Results:

• Hundreds of antibodies with strong biophysical properties were identified against tested antigens.
• Flow cytometry and immunohistochemistry validated the functionality of selected antibodies.
• Machine learning analysis identified additional antibodies for ROBO2 and PD-L2 from the generated dataset.

Conclusions:

• The developed synthetic library and associated dataset provide an ML-compatible framework for antibody discovery.
• This approach significantly streamlines the process of identifying and developing novel antibodies.
• The publicly available data facilitates further research and innovation in antibody engineering and therapeutic development.