Humanization of antibodies using a machine learning approach on large-scale repertoire data

Area of Science:

• Biotechnology
• Immunology
• Computational Biology

Background:

• Monoclonal antibody (mAb) therapeutics derived from non-human sources can trigger immunogenic responses in humans.
• Current humanization methods often involve extensive trial-and-error experimentation.
• Developing safe and effective human antibody therapeutics is crucial for clinical applications.

Purpose of the Study:

• To develop machine learning classifiers for distinguishing human and non-human antibody variable domain sequences.
• To create a computational tool, Hu-mAb, for predicting antibody humanization mutations.
• To reduce the time and experimental effort required for antibody humanization.

Main Methods:

• Trained machine learning classifiers on large antibody repertoire datasets.
• Validated classifier performance against existing models and experimental immunogenicity data.
• Developed the Hu-mAb tool to suggest sequence mutations for reduced immunogenicity.

Main Results:

• Classifiers demonstrated superior performance in discriminating human from murine sequences.
• Hu-mAb suggested mutations showed significant overlap with experimentally determined humanization changes.
• The computational approach offers a faster alternative to traditional humanization methods.

Conclusions:

• Machine learning offers a powerful approach to antibody sequence analysis and humanization.
• Hu-mAb effectively predicts mutations to reduce antibody immunogenicity.
• The Hu-mAb tool streamlines the development of safer, non-immunogenic antibody therapeutics.