VH-VL orientation prediction for antibody humanization candidate selection: A case study.

Area of Science:

• Immunology and Molecular Biology
• Biotechnology and Biopharmaceutical Development

Background:

• Antibody humanization grafts non-human antibody complementarity-determining regions onto human frameworks to reduce immunogenicity.
• Traditional humanization relies on sequence similarity for acceptor framework selection, potentially altering antigen-binding site geometry.

Purpose of the Study:

• To introduce a novel approach for antibody humanization focusing on preserving variable domain orientation.
• To improve the selection of acceptor frameworks by conserving the 3D geometry of the antigen-binding site.

Main Methods:

• Developed a machine learning-based predictor for antibody variable domain orientation.
• Selected acceptor frameworks based on conserved relative orientation between variable domains.
• Validated the approach using data from three humanization campaigns.

Main Results:

• Preselecting humanization variants based on predicted variable domain orientation significantly improved binding affinity.
• The novel method identified subsets of variants with enhanced antigen-binding capabilities compared to traditional methods.

Conclusions:

• Conserving variable domain orientation is a critical factor for successful antibody humanization.
• Machine learning-driven prediction of domain orientation offers a superior strategy for selecting humanization frameworks, leading to improved antibody efficacy.