Antibody-Antigen Docking Analysis
This tutorial demonstrates a complete AI-powered computational immunotherapy workflow using Chiral Potter's integrated platform. We'll predict antibody structures with Boltz-2, model complexes with DiffDock-PP, and analyze binding with comprehensive reporting and Mol* visualization.
Tutorial Overview
Learning Objectives
By completing this tutorial, you will:
- Predict antibody structures from sequence using Boltz-2 model
- Master protein-protein docking with DiffDock-PP
- Analyze binding interfaces with comprehensive docking reports and CAPRI metrics
- Create publication-quality visualizations with integrated Mol* viewer
Biological Context: Antibody-Antigen Recognition
We'll model the 4G6K complex interaction:
- System: Antibody-antigen binding complex from PDB 4G6K
- Application: Understanding molecular recognition mechanisms
- Research Goal: Analyze binding interfaces and interaction patterns
- Method: Complete AI-powered structure-to-complex workflow
Complete Workflow
Structure Prediction
Boltz-2 AI-powered antibody structure prediction
Protein-Protein Docking
DiffDock-PP complex formation modeling
Docking Report
Comprehensive binding analysis with CAPRI metrics
Mol* Visualization
Interactive 3D exploration and publication graphics
Stage 1: Structure Prediction with Boltz-2
Setting Up the Prediction
Create New Project and Workflow
- Project name: "4G6K-Complex-Analysis"
- Workflow name: "Antibody-antigen docking"
Prepare Sequence Input Upload the antibody sequences from the 4G6K complex:
Heavy chain (4G6K_H):
>A|protein
QVQLQESGPGLVKPSQTLSLTCSFSGFSLSTSGMGVGWIRQPSGKGLEWLAHIWWDGDESYN
PSLKSRLTISKDTSKNQVSLKITSVTAADTAVYFCARNRYDPPWFVDWGQGTLVTVSSASTK
GPSVFPLAPSSKSTSGGTAALGCLVKDYFPEPVTVSWNSGALTSGVHTFPAVLQSSGLYSLS
SVVTVPSSSLGTQTYICNVNHKPSNTKVDKRVEPLight chain (4G6K_L):
>B|protein
DIQMTQSTSSLSASVGDRVTITCRASQDISNYLSWYQQKPGKAVKLLIYYTSKLHSGVPSRF
SGSGSGTDYTLTISSLQQEDFATYFCLQGKMLPWTFGQGTKLEIKRTVAAPSVFIFPPSDEQ
LKSGTASVVCLLNNFYPREAKVQWKVDNALQSGNSQESVTEQDSKDSTYSLSSTLTLSKADY
EKHKVYACEVTHQGLSSPVTKSFNRGConfigure Boltz-2 Parameters
- Output Format: PDB
- Recycling Steps: 3
- Sampling Steps: 200
- Diffusion Samples: 1
- Step Scale: 1.638
- Devices: 1
- Accelerator: GPU
Understanding Boltz-2 Configuration
Input Options
Sequence Input
- FASTA format for antibody chains
- Multi-chain prediction for heavy and light chains
- MSA automatically generated
Quality Settings
Confidence Levels (0-1 scale)
- pLDDT > 0.9: Very high confidence
- pLDDT 0.7-0.9: Confident regions
- pLDDT 0.5-0.7: Low confidence
- pLDDT < 0.5: Very low confidence
Boltz-2 Confidence Score
Calculated as: 0.8 × overall pLDDT + 0.2 × interface pTM
- Combines local structure confidence with interface quality
- Range: 0 (low) to 1 (high confidence)
Antibody Features
Antibody-Specific Considerations
- CDR Loops: Complementarity-determining regions
- Framework Regions: Structural scaffold
- Heavy/Light Pairing: Chain association
- Antigen Binding: Paratope-epitope interaction
Boltz-2 Antibody Capabilities
- Trained on extensive antibody structures
- Handles CDR loop flexibility
- Predicts heavy-light chain orientation
- Models antigen binding interfaces
Stage 2: AI-Powered Protein-Protein Docking
Configuring DiffDock-PP
Add DiffDock-PP node to your workflow:
- num_samples: 40,
- run_name: "4G6K"
For antibody-antigen docking, provide the antibody as the ligand and antigen as the receptor for optimal results with DiffDock-PP's training data.
Docking Execution
The platform will:
Protein Encoding
- ESM-2 generates embeddings for both proteins
- Captures evolutionary and structural information
- Takes ~2 minutes for protein preparation
Pose Generation
- Diffusion model samples 60 binding orientations
- Explores rotational and translational space
- Focuses on biologically relevant interfaces
Quality Assessment
- Confidence scoring for each pose
DiffDock-PP Output Files
- Pickle files: Raw predictions and RMSD data
- PDB poses: Poses by ranking automatically exported for visualization
- Summary.txt: Job metadata and run information
Stage 3: Docking Report Generation
Understanding DiffDock-PP Output
The docking generates 40 poses with confidence scores:
Confidence Score Assessment
- High Confidence: score > 0
- Moderate Confidence: -1.5 < score ≤ 0
- Low Confidence: score ≤ -1.5
Additional Metrics (if reference available)
When a reference structure is provided, the system calculates Complex RMSD, Ligand RMSD, and Interface RMSD for detailed accuracy assessment.
The Docking Report processes DiffDock-PP output files to generate analysis:
Report Generation
- HTML dashboard: Interactive confidence analysis and pose ranking
- Quality assessment: High/Moderate/Low confidence classification
- Visualization: Confidence score plots and pose comparisons
Interactive Report Features
Dashboard Navigation
- Summary page: Key metrics and best poses at a glance
- Detailed analysis: Deep dive into individual poses
- Comparison tools: Side-by-side pose evaluation
- Export options: Download figures and data tables
Key Visualizations
- Score distributions: Statistical analysis of all poses
- Interface heatmaps: Interaction frequency across poses
- 3D overlay: Structural diversity visualization
- Interaction networks: Detailed molecular contacts
Stage 4: Mol* Visualization and Export
Integrated Mol* Viewer
The platform provides seamless embedded visualization:
Visualization Features
- Multi-pose comparison: Overlay and compare top solutions
- Interface analysis: Detailed interaction visualization
- Publication graphics: High-resolution image export
Conclusion
This tutorial demonstrates the power of fully AI-powered workflows for antibody-antigen analysis. By combining Boltz-2's structure prediction with DiffDock-PP's docking and comprehensive reporting tools, researchers can go from sequence to detailed binding analysis without requiring experimental structures.
Key Takeaways
- End-to-end AI enables structure-based drug design from sequence alone
- Comprehensive analysis through automated reporting and visualization
- Quality assessment using CAPRI metrics and multi-pose analysis
Resources and References
Key Publications
- Ille, A. M. et al. (2025). Human protein interactome structure prediction at scale with Boltz-2. bioRxiv. https://doi.org/10.1101/2025.07.03.663068
- Ketata, M. A. et al. (2023). DiffDock-PP: Rigid Protein-Protein Docking with Diffusion Models. arXiv:2304.03889.
- Sehnal, D. et al. (2021). Mol* Viewer: modern web app for 3D visualization and analysis of large biomolecular structures. Nucleic Acids Research, 49(W1), W431-W437.
- GitHub repositories:
- Boltz: https://github.com/jwohlwend/boltz
- DiffDock-PP: https://github.com/ketatam/DiffDock-PP
- Mol*: https://github.com/molstar/molstar
This tutorial demonstrates the complete AI-powered workflow for antibody-antigen docking analysis using Boltz-2, DiffDock-PP, and comprehensive reporting tools.