---
name: alphafold3-antibody-antigen-protocol
description: Predict antibody-antigen complex structures using AlphaFold 3, with specialized scoring for paratope-epitope interface quality.
allowed-tools: WebFetch, Bash(python *), Bash(mkdir *), Bash(cp *), Bash(ls *), Bash(jq *), Bash(cd *)
---

# AlphaFold 3 Antibody-Antigen Complex Predictor Protocol

## Purpose

Predict the binding mode between an antibody and its antigen target using AlphaFold 3. Focus on accurate paratope-epitope interface prediction with confidence scoring specific to antibody-antigen interactions.

## Inputs

Create an `inputs/` directory containing:

- `inputs/antibody.fasta` or `inputs/antibody.json`: Antibody sequence(s). Format as separate chains:
  - Heavy chain (VH): typically ~450-500 aa
  - Light chain (VL): typically ~210-220 aa
  - For nanobodies: single VHH domain
- `inputs/antigen.json`: AlphaFold 3 JSON for the antigen (protein, complex, or molecular system).
- `inputs/metadata.md`: 
  - Antibody name, origin (hybridoma, phage display, etc.)
  - Antigen name, known epitopes if known
  - Desired affinity if known
  - Experimental validation available (Y/N)

## Pre-Run Checks

1. Confirm research use is permitted.
2. Verify antibody sequences contain valid amino acid codes.
3. Check for signal peptides and clean to mature sequence if needed.
4. Verify antigen structure prediction is reasonable (good pLDDT) before complex prediction.
5. For glycosylated antigens, note that AF3 support is limited.

## Step 1: Prepare Antibody Input

For full antibodies, prepare as separate chains:

```json
{
  "name": "anti-TARGET antibody",
  "sequences": [
    {
      "protein_chain": {
        "sequence": "EVQLVESGGGLVQPGGSLRLSCAAS...",
        "id": {"value": "H"},
        "description": "Heavy chain variable domain"
      }
    },
    {
      "protein_chain": {
        "sequence": "DIQMTQSPSSLSASVGDRVTITC...",
        "id": {"value": "L"},
        "description": "Light chain variable domain"
      }
    }
  ]
}
```

For nanobodies (VHH), single chain is sufficient.

## Step 2: Validate Individual Components

### Antibody Validation

1. Check CDR regions are present (canonical motifs):
   - Heavy chain CDR1: positions ~31-35
   - Heavy chain CDR2: positions ~50-58
   - Heavy chain CDR3: positions ~95-110
2. Verify framework conserved residues (e.g., Cysteine at CDR positions for disulfide bonds)

### Antigen Validation

If antigen structure is already predicted or available:
1. Check pLDDT scores (should be > 70 for structured regions)
2. Note known epitope regions if available

## Step 3: Run Complex Prediction

### Route A: AlphaFold Server

1. Create new AlphaFold Server job.
2. Add antibody chain(s): first heavy chain, then light chain (or VHH alone).
3. Add antigen components.
4. Submit and wait for completion.
5. Download to `outputs/complex/`.

### Route B: Local AlphaFold 3

```bash
mkdir -p outputs/complex
python run_alphafold.py \
  --json_path=inputs/antibody_antigen.json \
  --output_dir=outputs/complex
```

## Step 4: Extract Paratope-Epitope Metrics

Analyze the predicted complex:

1. **Identify CDR residues**: Extract positions of CDR loops
2. **Identify epitope residues**: Residues on antigen within 5Å of any antibody atom
3. **Calculate interface metrics**:

```json
{
  "antibody": "anti-TARGET",
  "antigen": "TARGET protein",
  "interface_residues_antibody": [45, 46, 47, 98, 99, 100, 101],
  "interface_residues_antigen": [120, 121, 122, 150, 151, 152],
  "interface_pLDDT_antibody": 82.3,
  "interface_pLDDT_antigen": 78.5,
  "interface_pLDDT_mean": 80.4,
  "pae_interface_mean": 3.2,
  "contact_count": 45,
  "hydrogen_bonds_predicted": 8,
  "hydrophobic_contacts_predicted": 22,
  "interface_area_A2": 1450,
  "binding_confidence": "high"
}
```

## Step 5: Assess CDR Conformation Quality

1. Check if CDR loops have reasonable conformations (not overly distorted)
2. Compare CDR3 length to known distributions:
   - VH CDR3: typically 8-20 aa
   - VHH CDR3: typically 15-25 aa (longer than conventional antibodies)
3. Note if any CDR has very low pLDDT (< 60), indicating potential flexibility

## Step 6: Generate Report

Write `outputs/antibody_analysis.md`:

```markdown
# Antibody-Antigen Complex Prediction Report

## Antibody
- Name: [name]
- Type: [Full IgG / Nanobody / Fab / ScFv]
- Heavy chain length: [N] aa
- Light chain length: [N] aa
- CDR3 length: [N] aa

## Antigen
- Name: [name]
- Type: [Protein / Complex / Membrane protein]
- Length: [N] residues
- Source: [organism]

## Prediction Quality
- Overall pLDDT: [mean]
- Antigen pLDDT: [mean]
- Antibody pLDDT: [mean]

## Interface Analysis

### Paratope (Antibody binding site)
- Residues involved: [list]
- CDR contributions: CDR1: [N], CDR2: [N], CDR3: [N] residues
- Mean interface pLDDT: [value]

### Epitope (Antigen binding site)
- Residues involved: [list]
- Surface accessibility: [estimated]
- Mean interface pLDDT: [value]

### Interaction Quality
- Interface confidence: [High/Medium/Low]
- PAE at interface: [value] Å
- Predicted contacts: [N]
- Interface area: [N] Ų

## Binding Mode Assessment
[Describe the predicted binding orientation, whether it blocks known functional sites, etc.]

## Comparison with Known Data (if available)
- Predicted epitope vs. known epitope: [Consistent/Partial/No overlap/Unknown]
- Confidence implications: [note]

## Limitations
- AlphaFold 3 predictions are computational hypotheses
- CDR loops may be flexible and not fully captured
- Affinity/kinetics cannot be predicted from structure alone
- May not capture conformational changes upon binding
- Does not account for glycosylation effects

## Recommendations
1. Validate predicted epitope with mutagenesis (alanine scanning)
2. Test binding with SPR/BLI to confirm interaction
3. Consider cryo-EM or crystallography for definitive structure
4. If epitope is functional site, assess therapeutic implications

## References
- Dechaumes et al., AlphaFold2 for antibody modeling, MAbs, 2024
- AlphaFold 3: Abramson et al., Nature, 2024
```

## Success Criteria

- Antibody and antigen are correctly formatted and predicted.
- Interface metrics are extracted and quantified.
- CDR conformations are assessed for quality.
- Report provides actionable insights for experimental validation.
- Limitations are explicitly stated.

## Failure Modes

- Invalid antibody sequence (frameshift, unusual length) → validate sequence first
- Antigen prediction fails → improve antigen prediction before complex
- No predicted interface (chains separate) → may indicate no binding, retry or note
- Very low interface pLDDT → treat as uncertain, note limitation

## References

- Schrodinger, Antibody Structure Prediction with AlphaFold2, 2023
- AlphaFold 3: Abramson et al., Nature, 2024
- Kunstra et al., Protein-protein docking with antibody constraints, JACS, 2023