MVSF-AB: accurate antibody-antigen binding affinity prediction via multi-view sequence feature learning.

Area of Science:

• Biochemistry
• Immunology
• Computational Biology

Background:

• Accurate prediction of antibody-antigen binding affinity is critical for therapeutic antibody development, antibody engineering, and vaccine design.
• Traditional structure-based machine learning methods are limited by the high cost and unavailability of structural data for most antibodies and antigens.
• Existing sequence-based methods struggle with antibody-antigen affinity prediction due to imbalanced data and model designs not specific to antibody-antigen interactions.

Purpose of the Study:

• To develop an accurate sequence-based prediction method for antibody-antigen binding affinity.
• To address the limitations of existing methods in handling antibody-antigen interactions and imbalanced datasets.
• To propose a novel Multi-View Sequence Feature (MVSF-AB) learning approach.

Main Methods:

• Developed MVSF-AB, a multi-view learning method that integrates semantic and residue features from sequence data.
• Designed the model framework specifically for antibody-antigen interactions to capture key features.
• Utilized sequence information to predict binding affinity without relying on structural data.

Main Results:

• MVSF-AB effectively fuses multi-view sequence features for enhanced antibody-antigen affinity prediction.
• The proposed method demonstrates superior performance compared to existing approaches for predicting natural antibody-antigen affinity.
• MVSF-AB maintains prediction accuracy even when dealing with antibody mutations.

Conclusions:

• MVSF-AB offers a robust and accurate sequence-based solution for predicting antibody-antigen binding affinity.
• The method has significant implications for accelerating therapeutic antibody discovery and engineering.
• The approach overcomes limitations of structural data dependency and improves upon existing sequence-based prediction models.