ParaSurf is a deep learning-based approach for predicting paratope–antigen interactions by analyzing the molecular surface of antibodies. It integrates geometric, chemical, and electrostatic force-field features within a hybrid architecture of 3D ResNet and a transformer layer to enhance binding site prediction. Unlike previous models that focus only on the variable region, ParaSurf predicts binding scores across the entire Fab region, achieving state-of-the-art performance on major antibody–antigen benchmark datasets.

Example Use Cases:

• Predicting antibody binding sites for therapeutic antibody design.
• Facilitating vaccine development by identifying key antigen recognition sites.
• Enhancing structural bioinformatics research by improving paratope–antigen interaction prediction.
• Integrating with docking tools to refine antibody–antigen interaction modeling.

Technology:

• Deep learning model with 3D ResNet and transformer layers.
• Surface-based geometric, chemical, and electrostatic feature extraction.
• Trained on benchmark datasets: PECAN, Paragraph-expanded, and MIPE.
• Uses a voxelized 3D representation of molecular surfaces.

Limitations:

• Requires high-quality structural data for accurate predictions.
• Computationally intensive due to deep learning model complexity.