Interpretable Antibody-Antigen Structural Interface Prediction via Adaptive Graph Learning and Cyclic Transfer.

Area of Science:

• Structural Biology
• Immunology
• Computational Biology

Background:

• Experimental methods for identifying antibody-antigen interfaces are precise but time-consuming and resource-intensive.
• Computational prediction of these interfaces is crucial for accelerating antibody discovery and understanding immune recognition.
• Challenges include limited structural data, class imbalance, and complex biomolecular interactions.

Purpose of the Study:

• To develop a computational framework for accurate and efficient prediction of antibody-antigen interfaces.
• To accelerate antibody discovery and provide insights into the principles of immune recognition.

Main Methods:

• Introduction of VASCIF (Variable-domain Antibody-antigen Structural Complex Interface Finder), a structure-aware framework.
• Utilizing a Masked Graph Attention (MGA) architecture to represent protein complexes as residue graphs.
• Employing attention-based message passing to capture long-range structural dependencies.

Main Results:

• VASCIF achieves state-of-the-art performance in residue-level interface prediction on benchmark datasets.
• The framework demonstrates efficient inference, significantly faster than existing structure-based approaches.
• Interpretability analyses confirm the recovery of biophysically meaningful interaction patterns.

Conclusions:

• VASCIF provides a practical and efficient computational framework for predicting antibody-antigen interfaces.
• The model offers new insights into the molecular recognition principles governing antibody-antigen interactions.
• Redefining interfaces with larger distance thresholds improves predictive performance, highlighting the importance of broader interaction contexts.