Predicting Antibody Affinity Changes upon Mutation Based on Unbound Protein Structures.

Area of Science:

• Immunology
• Computational Biology
• Protein Engineering

Background:

• Antibodies are crucial for adaptive immunity, recognizing specific antigens.
• Monoclonal antibodies are valuable therapeutics, but improving their affinity is a key challenge.
• Current computational methods often require complex antigen-antibody structures, limiting their application.

Purpose of the Study:

• To develop a structure-free computational approach for predicting the impact of residue mutations on antibody affinity.
• To enable accurate affinity prediction for antibody engineering without experimental complex structures.

Main Methods:

• Utilized graph representation of proteins and a pre-trained encoder to capture residue microenvironments and antigen context.
• Developed a method that analyzes antibody mutations without requiring antigen-antibody complex structures.
• Curated a benchmark dataset specifically for antibody mutations.

Main Results:

• The developed approach achieves superior or comparable accuracy to structure-based and sequence-based methods on benchmark datasets.
• Demonstrated the method's effectiveness in predicting affinity changes for antibodies targeting SARS-CoV-2, influenza, and human cytomegalovirus.
• Validated the advantage of not requiring antigen-antibody complex structures for mutation effect prediction.

Conclusions:

• The novel computational method accurately predicts mutation effects on antibody affinity without complex structures.
• This approach has significant potential for practical antibody engineering and therapeutic development.
• Facilitates the identification of mutations that enhance antibody affinity in various applications.