---
name: peptide-virtual-screen-protocol
description: Comprehensive virtual screening pipeline for peptide-protein binding prediction using AlphaFold 3 structure prediction combined with binding interface analysis and multi-metric scoring.
allowed-tools: WebFetch, Bash(python *), Bash(mkdir *), Bash(cp *), Bash(ls *), Bash(jq *), Bash(cd *), Bash(cat *)
---

# Peptide Virtual Screening Pipeline

## Purpose

Screen candidate peptide sequences for binding to a target protein by predicting peptide-protein complex structures and scoring binding likelihood. This protocol transforms AlphaFold 3 structure predictions into a virtual screening platform suitable for computational prioritization of peptide libraries.

## Background and Motivation

Peptide-protein interactions mediate fundamental biological processes:
- Signal transduction and cellular communication
- Immune recognition and antibody-antigen binding
- Enzyme regulation and substrate recognition
- Gene regulation via transcription factor binding

Computational pre-screening can reduce experimental burden by prioritizing candidates most likely to bind. This protocol provides a reproducible, fully-documented workflow for peptide virtual screening.

## Scientific Foundation

This pipeline builds on established methods:

1. **AlphaFold 3** (Abramson et al., Nature, 2024): Provides accurate prediction of peptide-protein complex structures with built-in confidence metrics.

2. **CAMP Framework** (Ternavor et al., Nature Communications, 2021): Multi-level peptide-protein interaction prediction combining sequence and structural features.

3. **PepCNN** (Peterson et al., Scientific Reports, 2023): Deep learning for binding residue identification using protein language model embeddings.

## Input Files Specification

### Directory Structure

Create the following directory structure before running:

```
project/
????? inputs/
??  ????? target.json          # REQUIRED: Target protein in AF3 JSON format
??  ????? peptides.fasta       # REQUIRED: Peptide library in FASTA format
??  ????? screen_config.yaml   # REQUIRED: Scoring configuration
??  ????? peptides_metadata.md # OPTIONAL: Peptide annotations
??  ????? known_sites.txt      # OPTIONAL: Known binding residues on target
????? outputs/                 # Created automatically
```

### 1. Target Protein JSON (REQUIRED)

The target protein must be in AlphaFold 3 JSON format. Example:

```json
{
  "name": "Human_Hemoglobin_Alpha",
  "sequences": [
    {
      "protein": {
        "sequences": ["MVLSPADKTNVKAAWGKVGAHAGEYGAEALERMFLSFPTTKTYFPHFDLSH"],
        "count": 1
      }
    }
  ],
  "dialect": "alphafold",
  "version": 1
}
```

**Field Requirements:**
- `name`: String identifier for the target
- `sequences`: Array containing protein chain definitions
- `sequences[].protein.sequences`: Array with ONE sequence string (the target chain)
- `sequences[].protein.count`: Must be 1 for single-chain targets
- `dialect`: Must be "alphafold"
- `version`: Must be 1

**Multiple Chains:** If the target has multiple chains:
```json
{
  "name": "Target_Protein_Complex",
  "sequences": [
    {"protein": {"sequences": ["SEQ1..."], "count": 1}},
    {"protein": {"sequences": ["SEQ2..."], "count": 1}}
  ],
  "dialect": "alphafold",
  "version": 1
}
```

### 2. Peptide Library FASTA (REQUIRED)

Each peptide as a separate FASTA entry. Headers contain peptide identifiers.

```
>PEP_001 Human_proteome_derived
WLEAILPVGL
>PEP_002 Mouse_homolog_variant
WLEAILPVSL
>PEP_003 Alanine_scan_position_5
WLEAVLPVGL
>PEP_004 N_terminal_truncation
ALEAILPVGL
>PEP_005 Designed_variant_G10A
WLEAILPVGA
```

**Format Rules:**
- Header lines start with `>`
- Sequence lines contain only amino acid letters (A-Z, no whitespace except line breaks)
- Sequences should be 5-30 amino acids for optimal screening
- Valid amino acids: ACDEFGHIKLMNPQRSTVWY
- Avoid non-standard amino acids (X, B, Z, U, O) unless using local AlphaFold 3 with custom handling

### 3. Screening Configuration YAML (REQUIRED)

```yaml
# inputs/screen_config.yaml

# === Screening Parameters ===
screen:
  max_peptide_length: 30      # Maximum peptide length to screen
  min_peptide_length: 5       # Minimum peptide length to screen
  batch_size: 50              # Number of predictions per batch
  max_predictions_per_run: 500 # Safety limit for predictions

# === Scoring Weights ===
# These weights sum to 1.0 (100%)
# Adjust based on your target characteristics
scoring:
  interface_pLDDT_weight: 0.35      # Weight for interface confidence
  pae_interchain_weight: 0.25      # Weight for positional accuracy
  contact_count_weight: 0.20        # Weight for interaction extent
  peptide_length_bonus: 0.20        # Weight for length suitability

# === Confidence Thresholds ===
# Scores are on 0-100 scale
thresholds:
  high_confidence: 75    # >= 75: Strong binding evidence
  medium_confidence: 55  # 55-74: Moderate evidence
  low_confidence: 0     # < 55: Weak/no predicted binding

# === Interface Detection ===
interface:
  distance_threshold_angstrom: 5.0  # Distance for defining interface residues
  min_interface_residues: 3         # Minimum interface residues required

# === Output Options ===
output:
  top_n_for_report: 10               # Number of top candidates in report
  save_all_scores: true              # Save scores for all peptides
  save_predictions: false            # Whether to save full AF3 predictions
  verbose_logging: true             # Enable detailed logging
```

### 4. Peptide Metadata (OPTIONAL)

Provide additional context for each peptide:

```markdown
# inputs/peptides_metadata.md

## PEP_001
- **Source**: Human proteome
- **Known activity**: Binds hemoglobin with Kd ~ 100nM
- **Modifications**: None
- **Notes**: Reference peptide for comparison

## PEP_002
- **Source**: Mouse homolog
- **Known activity**: Unknown
- **Modifications**: Single Ser substitution
- **Notes**: Position 10 variation

## PEP_003
- **Source**: Alanine scan library
- **Notes**: Systematic alanine substitution at position 5
```

### 5. Known Binding Sites (OPTIONAL)

If known binding sites exist on the target:

```
# inputs/known_sites.txt
# Format: RESIDUE_NUMBER CHAIN (one per line)
# Example:
15 A
16 A
17 A
20 A
21 A
```

## Scoring System Details

### Metric Definitions

**1. Interface pLDDT (35% weight)**
- What: Mean per-residue confidence score for interface residues
- Range: 0-100
- Interpretation: Higher is better (>90 = very confident, >70 = confident, <50 = low confidence)
- Source: AlphaFold 3 confidence JSON

**2. Inter-chain PAE (25% weight)**
- What: Position-specific error estimate between peptide and protein chains
- Range: 0 to ~30+ Angstroms
- Interpretation: Lower is better (<5 = accurate positioning, 5-10 = uncertain, >15 = poor)
- Source: AlphaFold 3 PAE matrix, inter-chain block

**3. Contact Count (20% weight)**
- What: Number of residue pairs across chains within distance threshold
- Range: 0 to unlimited (typically 0-50 for peptide-protein)
- Interpretation: More contacts suggest more extensive interface
- Source: Parsed from predicted complex structure

**4. Length Suitability (20% weight)**
- What: Score based on typical peptide lengths for binding
- Range: 0-100
- Optimal: 8-20 amino acids
- Acceptable: 5-7 or 21-30 amino acids
- Suboptimal: < 5 or > 30 amino acids

### Composite Score Formula

```
composite_score = (interface_pLDDT_normalized ? 0.35)
                + (PAE_normalized ? 0.25)
                + (contacts_normalized ? 0.20)
                + (length_score ? 0.20)
```

Where:
- `interface_pLDDT_normalized` = min(100, max(0, interface_plddt))
- `PAE_normalized` = max(0, 100 - pae ? 5)
- `contacts_normalized` = min(100, contacts ? 5)
- `length_score` = 100 if 8-20 aa, 80 if 5-7 or 21-30 aa, else 50

### Confidence Categories

| Category | Score Range | Interpretation | Action |
|----------|-------------|----------------|--------|
| High | >= 75 | Strong computational evidence | Prioritize for experimental validation |
| Medium | 55-74 | Moderate evidence | Include in validation panel |
| Low | < 55 | Weak/no predicted binding | Defer unless resources available |

## Step-by-Step Protocol

### Step 1: Environment Setup

```bash
# Create and activate environment
python -m venv venv
source venv/bin/activate  # Linux/Mac
# venv\Scripts\activate   # Windows

# Install dependencies
pip install biopython pyyaml numpy

# Verify installations
python -c "import Bio; import yaml; import numpy; print('Dependencies OK')"
```

### Step 2: Parse Input Data

Create `scripts/parse_inputs.py`:

```python
#!/usr/bin/env python3
"""
Parse input files and create manifest for screening.
"""
import json
import yaml
from Bio import SeqIO
from pathlib import Path

def parse_inputs(inputs_dir="inputs", outputs_dir="outputs"):
    """Parse all input files and create manifest."""
    inputs_path = Path(inputs_dir)
    outputs_path = Path(outputs_dir)
    outputs_path.mkdir(parents=True, exist_ok=True)

    # Parse target JSON
    target_path = inputs_path / "target.json"
    if not target_path.exists():
        raise FileNotFoundError(f"Target JSON not found: {target_path}")
    with open(target_path) as f:
        target = json.load(f)

    # Validate target format
    validate_target_format(target)

    # Parse peptides FASTA
    peptides_path = inputs_path / "peptides.fasta"
    if not peptides_path.exists():
        raise FileNotFoundError(f"Peptides FASTA not found: {peptides_path}")

    peptides = []
    invalid_sequences = []

    for record in SeqIO.parse(str(peptides_path), "fasta"):
        seq = str(record.seq).upper()
        peptide_id = record.id

        # Validate sequence
        if not validate_peptide_sequence(seq):
            invalid_sequences.append((peptide_id, seq))
            continue

        # Check length constraints
        length = len(seq)
        peptides.append({
            'id': peptide_id,
            'name': record.description,
            'sequence': seq,
            'length': length
        })

    if invalid_sequences:
        print(f"Warning: {len(invalid_sequences)} peptides skipped due to invalid characters")
        for pid, seq in invalid_sequences:
            print(f"  - {pid}: {seq}")

    # Parse config if exists
    config_path = inputs_path / "screen_config.yaml"
    if config_path.exists():
        with open(config_path) as f:
            config = yaml.safe_load(f)
    else:
        config = get_default_config()

    # Create manifest
    manifest = {
        'target': target,
        'target_name': target.get('name', 'Unknown'),
        'peptides': peptides,
        'total_peptides': len(peptides),
        'config': config,
        'parse_info': {
            'input_dir': str(inputs_path.absolute()),
            'output_dir': str(outputs_path.absolute())
        }
    }

    # Save manifest
    manifest_path = outputs_path / "manifest.json"
    with open(manifest_path, 'w') as f:
        json.dump(manifest, f, indent=2)

    print(f"Parsed {len(peptides)} valid peptides from target {manifest['target_name']}")
    print(f"Manifest saved to: {manifest_path}")

    return manifest

def validate_target_format(target):
    """Validate AlphaFold 3 JSON format."""
    required_fields = ['sequences', 'dialect', 'version']
    for field in required_fields:
        if field not in target:
            raise ValueError(f"Target JSON missing required field: {field}")

    if target.get('dialect') != 'alphafold':
        raise ValueError(f"Target dialect must be 'alphafold', got: {target.get('dialect')}")

    if not target.get('sequences') or len(target['sequences']) == 0:
        raise ValueError("Target must have at least one sequence in 'sequences' array")

    return True

def validate_peptide_sequence(sequence):
    """Check if sequence contains only standard amino acids."""
    valid_amino_acids = set('ACDEFGHIKLMNPQRSTVWY')
    sequence_upper = sequence.upper()
    return all(aa in valid_amino_acids for aa in sequence_upper if aa not in ' \t\n')

def get_default_config():
    """Return default screening configuration."""
    return {
        'screen': {
            'max_peptide_length': 30,
            'min_peptide_length': 5,
            'batch_size': 50
        },
        'scoring': {
            'interface_pLDDT_weight': 0.35,
            'pae_interchain_weight': 0.25,
            'contact_count_weight': 0.20,
            'peptide_length_bonus': 0.20
        },
        'thresholds': {
            'high_confidence': 75,
            'medium_confidence': 55,
            'low_confidence': 0
        },
        'interface': {
            'distance_threshold_angstrom': 5.0,
            'min_interface_residues': 3
        },
        'output': {
            'top_n_for_report': 10,
            'save_all_scores': True,
            'verbose_logging': True
        }
    }

if __name__ == "__main__":
    import argparse
    parser = argparse.ArgumentParser(description="Parse peptide screening inputs")
    parser.add_argument("--inputs", default="inputs", help="Input directory")
    parser.add_argument("--outputs", default="outputs", help="Output directory")
    args = parser.parse_args()
    parse_inputs(args.inputs, args.outputs)
```

### Step 3: Predict Peptide-Protein Complexes

#### Option A: AlphaFold Server (Recommended for Small Libraries)

1. Go to: https://alphafold.ebi.ac.uk
2. Select "Complex Structure Prediction"
3. Upload target protein sequence first
4. Add each peptide as an additional chain
5. Submit job
6. Wait for completion (~10-15 minutes)
7. Download all results to `outputs/predictions/<peptide_id>/`

**Important:** AlphaFold Server has rate limits. For large libraries, consider:
- Using local AlphaFold 3 installation
- Batch submissions with delays
- Using institutional AlphaFold batch API

#### Option B: Local AlphaFold 3 Installation

```bash
# Create prediction scripts for each peptide
mkdir -p outputs/predictions

# Example batch script (adapt to your AF3 installation)
for peptide_id in PEP_001 PEP_002 PEP_003; do
  python run_alphafold.py \
    --json_path="inputs/complex_${peptide_id}.json" \
    --output_dir="outputs/predictions/${peptide_id}" \
    --model_preset=multimer
done
```

**Required AF3 JSON format for binary complex:**
```json
{
  "name": "Target_PEP_001_Complex",
  "sequences": [
    {"protein": {"sequences": ["TARGET_SEQUENCE..."], "count": 1}},
    {"protein": {"sequences": ["WLEAILPVGL"], "count": 1}}
  ],
  "dialect": "alphafold",
  "version": 1
}
```

### Step 4: Extract Binding Metrics

Create `scripts/extract_metrics.py`:

```python
#!/usr/bin/env python3
"""
Extract binding-relevant metrics from AlphaFold 3 predictions.
"""
import json
import numpy as np
from pathlib import Path
from Bio.PDB import MMCIFParser
import argparse

def extract_metrics(prediction_dir, peptide_seq, config):
    """
    Extract binding metrics from AlphaFold 3 prediction results.

    Args:
        prediction_dir: Path to prediction output directory
        peptide_seq: Peptide sequence string
        config: Configuration dict

    Returns:
        dict: Extracted metrics
    """
    pred_path = Path(prediction_dir)

    # Load confidence metrics
    conf_path = pred_path / "summary_confidences.json"
    if not conf_path.exists():
        raise FileNotFoundError(f"Confidence file not found: {conf_path}")

    with open(conf_path) as f:
        conf = json.load(f)

    # Load structure file for contact analysis
    structure_path = find_structure_file(pred_path)
    if structure_path is None:
        raise FileNotFoundError(f"No structure file found in {prediction_dir}")

    # Extract metrics
    metrics = {
        'prediction_dir': str(pred_path),
        'mean_plddt': conf.get('mean_plddt', 0),
        'mean_plddt_per_chain': conf.get('mean_plddt_per_chain', []),
        'ptm': conf.get('ptm', 0),
        'interface_plddt': extract_interface_plddt(conf),
        'pae_interchain': extract_interchain_pae(conf, len(peptide_seq)),
        'contact_count': calculate_contacts(structure_path, peptide_seq, config),
        'peptide_length': len(peptide_seq)
    }

    return metrics

def find_structure_file(pred_path):
    """Find the predicted structure file (CIF or PDB format)."""
    for ext in ['.cif', '.pdb']:
        files = list(pred_path.glob(f"*{ext}"))
        if files:
            return files[0]
    return None

def extract_interface_plddt(conf):
    """
    Extract pLDDT scores for interface residues.

    AlphaFold 3 may provide per-residue confidence. If available,
    use the mean pLDDT for residues identified as interface.
    """
    # Look for interface-specific pLDDT if available
    if 'interface_plddt_mean' in conf:
        return conf['interface_plddt_mean']

    # Fallback: use overall mean pLDDT as approximation
    if 'mean_plddt' in conf:
        return conf['mean_plddt']

    return 0

def extract_interchain_pae(conf, peptide_length):
    """
    Extract inter-chain PAE for peptide-protein interface.

    PAE (Predicted Alignment Error) measures the expected error
    in the relative position between residues. Lower values indicate
    higher confidence in the predicted positioning.
    """
    # PAE matrix if available
    if 'pae' in conf:
        pae_matrix = np.array(conf['pae'])

        # Inter-chain PAE: block corresponding to peptide-protein interface
        # Assuming peptide is the last chain
        n_protein = len(pae_matrix) - peptide_length

        if n_protein > 0 and len(pae_matrix) > peptide_length:
            # Extract inter-chain region
            interchain_pae = pae_matrix[:n_protein, n_protein:]
            return float(np.mean(interchain_pae))

    # Fallback to overall PAE if inter-chain not available
    if 'pae_interchain_mean' in conf:
        return conf['pae_interchain_mean']

    return 99  # Default high error if not available

def calculate_contacts(structure_path, peptide_seq, config):
    """
    Calculate number of contacts between peptide and protein.

    Contacts are defined as residue pairs where any atom pair
    is within the distance threshold (default 5 Angstrom).
    """
    parser = MMCIFParser(QUIET=True)
    structure = parser.get_structure('complex', str(structure_path))

    # Identify chains (assume last chain is peptide)
    chains = list(structure[0].get_chains())
    if len(chains) < 2:
        return 0

    peptide_chain = chains[-1]
    protein_chains = chains[:-1]

    distance_threshold = config.get('interface', {}).get('distance_threshold_angstrom', 5.0)

    # Get peptide residues
    peptide_residues = list(peptide_chain.get_residues())

    # Find contacts with protein chains
    contacts = set()

    for pep_res in peptide_residues:
        pep_atoms = list(pep_res.get_atoms())

        for prot_chain in protein_chains:
            for prot_res in prot_chain.get_residues():
                prot_atoms = list(prot_res.get_atoms())

                # Check if any atom pair is within threshold
                for pa in pep_atoms:
                    for ta in prot_atoms:
                        if pa - ta < distance_threshold:
                            contacts.add((pep_res.get_id(), prot_res.get_id()))
                            break  # One contact per residue pair is enough

    return len(contacts)

def process_all_predictions(manifest_path, config):
    """Process all predictions in manifest."""
    with open(manifest_path) as f:
        manifest = json.load(f)

    peptides = manifest['peptides']
    predictions_dir = Path(manifest['parse_info']['output_dir']) / 'predictions'

    results = []
    for peptide in peptides:
        peptide_id = peptide['id']
        seq = peptide['sequence']

        pred_dir = predictions_dir / peptide_id
        if not pred_dir.exists():
            print(f"Warning: Prediction not found for {peptide_id}")
            continue

        try:
            metrics = extract_metrics(pred_dir, seq, config)
            metrics['peptide_id'] = peptide_id
            results.append(metrics)
            print(f"Extracted metrics for {peptide_id}: contacts={metrics['contact_count']}, pLDDT={metrics['interface_plddt']:.1f}")
        except Exception as e:
            print(f"Error processing {peptide_id}: {e}")

    return results

if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="Extract metrics from AF3 predictions")
    parser.add_argument("--manifest", default="outputs/manifest.json")
    parser.add_argument("--config", default="inputs/screen_config.yaml")
    args = parser.parse_args()

    # Load config
    import yaml
    with open(args.config) as f:
        config = yaml.safe_load(f)

    results = process_all_predictions(args.manifest, config)

    # Save results
    with open("outputs/metrics.json", 'w') as f:
        json.dump(results, f, indent=2)

    print(f"Saved metrics for {len(results)} predictions")
```

### Step 5: Calculate Binding Scores

Create `scripts/calculate_scores.py`:

```python
#!/usr/bin/env python3
"""
Calculate composite binding scores from extracted metrics.
"""
import json
import yaml
from pathlib import Path
import argparse

def calculate_binding_score(metrics, config):
    """
    Calculate composite binding likelihood score.

    The score combines:
    - Interface pLDDT (35%): Confidence in the predicted interface
    - Inter-chain PAE (25%): Positional accuracy between chains
    - Contact count (20%): Extent of physical interaction
    - Length suitability (20%): Appropriateness of peptide length
    """
    weights = config['scoring']

    # 1. Interface pLDDT (already 0-100)
    plddt_score = normalize_plddt(metrics.get('interface_plddt', 0))

    # 2. PAE (invert so lower is better)
    pae_score = normalize_pae(metrics.get('pae_interchain', 99))

    # 3. Contact count (normalize to 0-100)
    contact_score = normalize_contacts(metrics.get('contact_count', 0))

    # 4. Length suitability
    length_score = normalize_length(metrics.get('peptide_length', 10))

    # Composite score
    composite = (
        plddt_score * weights['interface_pLDDT_weight'] +
        pae_score * weights['pae_interchain_weight'] +
        contact_score * weights['contact_count_weight'] +
        length_score * weights['peptide_length_bonus']
    )

    # Determine confidence category
    thresholds = config['thresholds']
    if composite >= thresholds['high_confidence']:
        category = 'high'
    elif composite >= thresholds['medium_confidence']:
        category = 'medium'
    else:
        category = 'low'

    return {
        'peptide_id': metrics['peptide_id'],
        'composite_score': round(composite, 2),
        'confidence_category': category,
        'interface_plddt': round(plddt_score, 2),
        'pae_interchain': round(pae_score, 2),
        'contact_count': metrics.get('contact_count', 0),
        'contact_score': round(contact_score, 2),
        'length_score': round(length_score, 2),
        'component_scores': {
            'interface_confidence': round(plddt_score, 2),
            'positional_accuracy': round(pae_score, 2),
            'contact_extent': round(contact_score, 2),
            'length_suitability': round(length_score, 2)
        },
        'raw_metrics': {
            'interface_plddt_raw': metrics.get('interface_plddt', 0),
            'pae_interchain_raw': metrics.get('pae_interchain', 99),
            'contact_count_raw': metrics.get('contact_count', 0),
            'peptide_length': metrics.get('peptide_length', 0)
        }
    }

def normalize_plddt(plddt):
    """Normalize pLDDT to 0-100 scale."""
    return min(100, max(0, plddt))

def normalize_pae(pae):
    """
    Normalize PAE to 0-100 scale (inverted).

    PAE typically ranges 0-30+. Lower is better.
    We convert: 0 PAE -> 100 score, 20+ PAE -> 0 score
    """
    return max(0, 100 - pae * 5)

def normalize_contacts(n_contacts):
    """
    Normalize contact count to 0-100 scale.

    We use a simple scaling where ~20 contacts = 100 score.
    """
    return min(100, n_contacts * 5)

def normalize_length(length):
    """
    Score peptide length suitability.

    - 8-20 aa: Optimal (100)
    - 5-7 or 21-30 aa: Acceptable (80)
    - < 5 or > 30 aa: Suboptimal (50)
    """
    if 8 <= length <= 20:
        return 100
    elif 5 <= length < 8 or 20 < length <= 30:
        return 80
    else:
        return 50

def score_all_predictions(metrics_path, config):
    """Calculate scores for all predictions."""
    with open(metrics_path) as f:
        metrics_list = json.load(f)

    scores = []
    for metrics in metrics_list:
        score = calculate_binding_score(metrics, config)
        scores.append(score)

    # Sort by composite score
    ranked = sorted(scores, key=lambda x: x['composite_score'], reverse=True)

    # Add rank
    for i, s in enumerate(ranked, 1):
        s['rank'] = i

    return ranked

def generate_rankings(scores, manifest_path):
    """Generate final rankings JSON."""
    with open(manifest_path) as f:
        manifest = json.load(f)

    # Count by category
    high = sum(1 for s in scores if s['confidence_category'] == 'high')
    medium = sum(1 for s in scores if s['confidence_category'] == 'medium')
    low = sum(1 for s in scores if s['confidence_category'] == 'low')

    # Top priorities for validation
    priorities = [s['peptide_id'] for s in scores[:5] if s['confidence_category'] != 'low']

    rankings = {
        'screened_on': manifest.get('parse_info', {}).get('timestamp', ''),
        'target': manifest.get('target_name', 'Unknown'),
        'total_peptides': manifest['total_peptides'],
        'completed': len(scores),
        'high_confidence_count': high,
        'medium_confidence_count': medium,
        'low_confidence_count': low,
        'rankings': scores,
        'priorities_for_validation': priorities
    }

    return rankings

if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="Calculate binding scores")
    parser.add_argument("--metrics", default="outputs/metrics.json")
    parser.add_argument("--manifest", default="outputs/manifest.json")
    parser.add_argument("--config", default="inputs/screen_config.yaml")
    args = parser.parse_args()

    with open(args.config) as f:
        config = yaml.safe_load(f)

    scores = score_all_predictions(args.metrics, config)
    rankings = generate_rankings(scores, args.manifest)

    # Save
    with open("outputs/rankings.json", 'w') as f:
        json.dump(rankings, f, indent=2)

    # Print summary
    print(f"Scored {len(scores)} peptides:")
    print(f"  High confidence: {rankings['high_confidence_count']}")
    print(f"  Medium confidence: {rankings['medium_confidence_count']}")
    print(f"  Low confidence: {rankings['low_confidence_count']}")
    print(f"\nTop 5 priorities: {rankings['priorities_for_validation']}")
```

### Step 6: Generate Screening Report

Create `scripts/generate_report.py`:

```python
#!/usr/bin/env python3
"""
Generate formatted screening report.
"""
import json
from datetime import datetime
from pathlib import Path
import argparse

def generate_screen_report(rankings_path, output_path="outputs/screen_report.md"):
    """Generate comprehensive screening report."""

    with open(rankings_path) as f:
        rankings = json.load(f)

    report_lines = []

    # Header
    report_lines.append("# Peptide Virtual Screening Report")
    report_lines.append("")
    report_lines.append(f"**Generated:** {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}")
    report_lines.append(f"**Target:** {rankings['target']}")
    report_lines.append("")

    # Summary Statistics
    report_lines.append("## Summary Statistics")
    report_lines.append("")
    report_lines.append(f"| Metric | Value |")
    report_lines.append(f"|--------|-------|")
    report_lines.append(f"| Total peptides screened | {rankings['total_peptides']} |")
    report_lines.append(f"| Predictions completed | {rankings['completed']} |")
    report_lines.append(f"| High confidence binders | {rankings['high_confidence_count']} |")
    report_lines.append(f"| Medium confidence candidates | {rankings['medium_confidence_count']} |")
    report_lines.append(f"| Low confidence / non-binders | {rankings['low_confidence_count']} |")
    report_lines.append("")

    # Confidence Distribution
    if rankings['completed'] > 0:
        high_pct = rankings['high_confidence_count'] / rankings['completed'] * 100
        med_pct = rankings['medium_confidence_count'] / rankings['completed'] * 100
        low_pct = rankings['low_confidence_count'] / rankings['completed'] * 100
        report_lines.append(f"**Distribution:** {high_pct:.1f}% high, {med_pct:.1f}% medium, {low_pct:.1f}% low")
        report_lines.append("")

    # Top Candidates Table
    report_lines.append("## Top Candidates (Full Ranking)")
    report_lines.append("")
    report_lines.append("| Rank | Peptide ID | Score | Confidence | Interface pLDDT | PAE | Contacts |")
    report_lines.append("|------|------------|-------|------------|-----------------|-----|----------|")

    for r in rankings['rankings'][:20]:  # Top 20
        raw = r.get('raw_metrics', {})
        report_lines.append(
            f"| {r['rank']} | {r['peptide_id']} | {r['composite_score']} | "
            f"{r['confidence_category']} | {raw.get('interface_plddt_raw', 'N/A')} | "
            f"{raw.get('pae_interchain_raw', 'N/A')} | {raw.get('contact_count_raw', 'N/A')} |"
        )
    report_lines.append("")

    # Prioritized Validation List
    report_lines.append("## Recommended for Experimental Validation")
    report_lines.append("")
    if rankings['priorities_for_validation']:
        for i, pid in enumerate(rankings['priorities_for_validation'], 1):
            # Find the entry
            entry = next((r for r in rankings['rankings'] if r['peptide_id'] == pid), None)
            if entry:
                report_lines.append(f"{i}. **{pid}** - Composite Score: {entry['composite_score']} ({entry['confidence_category']} confidence)")
                report_lines.append(f"   - Interface pLDDT: {entry['raw_metrics'].get('interface_plddt_raw', 'N/A')}")
                report_lines.append(f"   - Inter-chain PAE: {entry['raw_metrics'].get('pae_interchain_raw', 'N/A')}")
                report_lines.append(f"   - Contact Count: {entry['raw_metrics'].get('contact_count_raw', 'N/A')}")
    else:
        report_lines.append("No high-confidence candidates identified.")
    report_lines.append("")

    # Detailed Scoring Explanation
    report_lines.append("## Scoring Methodology")
    report_lines.append("")
    report_lines.append("### Composite Score Components")
    report_lines.append("")
    report_lines.append("| Component | Weight | Description |")
    report_lines.append("|-----------|--------|-------------|")
    report_lines.append("| Interface pLDDT | 35% | Per-residue confidence at the peptide-protein interface |")
    report_lines.append("| Inter-chain PAE | 25% | Position-specific error between chains (lower is better) |")
    report_lines.append("| Contact Count | 20% | Number of residue pairs within 5 Angstrom |")
    report_lines.append("| Length Suitability | 20% | Optimal length: 8-20 amino acids |")
    report_lines.append("")

    report_lines.append("### Confidence Categories")
    report_lines.append("")
    report_lines.append("| Category | Score Range | Interpretation |")
    report_lines.append("|----------|-------------|----------------|")
    report_lines.append("| High | >= 75 | Strong computational evidence for binding |")
    report_lines.append("| Medium | 55-74 | Moderate evidence; validation recommended |")
    report_lines.append("| Low | < 55 | Weak or no predicted binding |")
    report_lines.append("")

    # Limitations
    report_lines.append("## Limitations and Caveats")
    report_lines.append("")
    report_lines.append("1. **Computational predictions are hypotheses**, not experimental evidence")
    report_lines.append("2. **AlphaFold 3 limitations:**")
    report_lines.append("   - May struggle with very short peptides (< 5 residues)")
    report_lines.append("   - Membrane proteins and disordered regions remain challenging")
    report_lines.append("   - Does not account for post-translational modifications")
    report_lines.append("3. **Does not predict binding affinity** (Kd), only binding likelihood")
    report_lines.append("4. **Missing cellular context:**")
    report_lines.append("   - Cellular concentrations")
    report_lines.append("   - Competitive binders")
    report_lines.append("   - Allosteric effects")
    report_lines.append("")

    # Recommendations
    report_lines.append("## Experimental Validation Recommendations")
    report_lines.append("")
    report_lines.append("### Recommended Methods")
    report_lines.append("")
    report_lines.append("| Method | Measures | Throughput | Notes |")
    report_lines.append("|--------|----------|------------|-------|")
    report_lines.append("| Surface Plasmon Resonance (SPR) | Kd, kon, koff | Low | Gold standard for binding kinetics |")
    report_lines.append("| Isothermal Titration Calorimetry (ITC) | Kd, ?H, ?S | Very Low | Direct enthalpy measurement |")
    report_lines.append("| Fluorescence Polarization (FP) | Kd | Medium | High-throughput screening |")
    report_lines.append("| AlphaScreen/ALPHALISA | Binding | High | Suitable for large panels |")
    report_lines.append("| Co-immunoprecipitation | Complex formation | Medium | Endogenous context |")
    report_lines.append("")

    report_lines.append("### Follow-up Studies")
    report_lines.append("")
    report_lines.append("- **Alanine scanning**: Identify key binding residues")
    report_lines.append("- **Specificity profiling**: Test against related proteins")
    report_lines.append("- **Stability assays**: Thermal shift, proteolytic resistance")
    report_lines.append("- **Cellular activity**: Functional assays if relevant")
    report_lines.append("")

    # References
    report_lines.append("## References")
    report_lines.append("")
    report_lines.append("1. Abramson et al. 'Accurate structure prediction of biomolecular interactions with AlphaFold 3.' Nature, 2024.")
    report_lines.append("2. Ternavor et al. 'A deep-learning framework for multi-level peptide-protein interaction prediction.' Nature Communications, 2021.")
    report_lines.append("3. Peterson et al. 'PepCNN deep learning tool for predicting peptide binding residues.' Scientific Reports, 2023.")
    report_lines.append("")

    report_content = "\n".join(report_lines)

    with open(output_path, 'w') as f:
        f.write(report_content)

    print(f"Report saved to: {output_path}")
    return report_content

if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="Generate screening report")
    parser.add_argument("--rankings", default="outputs/rankings.json")
    parser.add_argument("--output", default="outputs/screen_report.md")
    args = parser.parse_args()

    generate_screen_report(args.rankings, args.output)
```

### Step 7: Main Pipeline Script

Create `scripts/run_screen.py`:

```python
#!/usr/bin/env python3
"""
Main pipeline script for peptide virtual screening.
Run all steps: parse -> predict -> extract -> score -> report
"""
import subprocess
import sys
from pathlib import Path

def run_command(cmd, description):
    """Run a command and handle errors."""
    print(f"\n{'='*60}")
    print(f"STEP: {description}")
    print(f"CMD: {cmd}")
    print('='*60)
    result = subprocess.run(cmd, shell=True)
    if result.returncode != 0:
        print(f"ERROR: {description} failed with code {result.returncode}")
        sys.exit(1)
    print(f"SUCCESS: {description}")

def main():
    # Create necessary directories
    Path("inputs").mkdir(exist_ok=True)
    Path("outputs").mkdir(exist_ok=True)
    Path("outputs/predictions").mkdir(exist_ok=True)
    Path("scripts").mkdir(exist_ok=True)

    print("Peptide Virtual Screening Pipeline")
    print("="*60)

    # Step 1: Parse inputs
    run_command(
        "python scripts/parse_inputs.py --inputs inputs --outputs outputs",
        "Parse Input Data"
    )

    # Step 2: Predict complexes (manual step - requires AlphaFold 3)
    print("\n" + "="*60)
    print("STEP: AlphaFold 3 Predictions")
    print("="*60)
    print("NOTE: This step requires manual execution using AlphaFold 3")
    print("Options:")
    print("  1. AlphaFold Server: https://alphafold.ebi.ac.uk")
    print("  2. Local AlphaFold 3 installation")
    print("Place prediction results in: outputs/predictions/<peptide_id>/")
    print("="*60)

    # Step 3: Extract metrics
    run_command(
        "python scripts/extract_metrics.py --manifest outputs/manifest.json --config inputs/screen_config.yaml",
        "Extract Binding Metrics"
    )

    # Step 4: Calculate scores
    run_command(
        "python scripts/calculate_scores.py --metrics outputs/metrics.json --manifest outputs/manifest.json --config inputs/screen_config.yaml",
        "Calculate Binding Scores"
    )

    # Step 5: Generate report
    run_command(
        "python scripts/generate_report.py --rankings outputs/rankings.json --output outputs/screen_report.md",
        "Generate Screening Report"
    )

    print("\n" + "="*60)
    print("PIPELINE COMPLETE")
    print("="*60)
    print("Results:")
    print("  - Rankings: outputs/rankings.json")
    print("  - Report: outputs/screen_report.md")
    print("="*60)

if __name__ == "__main__":
    main()
```

## Output Files

After running the pipeline, the following files are generated:

```
outputs/
????? manifest.json           # Parsed input manifest
????? metrics.json            # Extracted metrics for all predictions
????? rankings.json          # Ranked candidates with scores
????? screen_report.md       # Human-readable report
????? predictions/           # AlphaFold 3 prediction results
    ????? PEP_001/
    ??  ????? summary_confidences.json
    ??  ????? predicted_structure.cif
    ????? PEP_002/
    ????? ...
```

### rankings.json Schema

```json
{
  "screened_on": "2026-04-30",
  "target": "Target_Protein_Name",
  "total_peptides": 100,
  "completed": 100,
  "high_confidence_count": 15,
  "medium_confidence_count": 35,
  "low_confidence_count": 50,
  "rankings": [
    {
      "rank": 1,
      "peptide_id": "PEP_042",
      "composite_score": 87.3,
      "confidence_category": "high",
      "interface_plddt": 91.2,
      "pae_interchain": 3.8,
      "contact_count": 18,
      "component_scores": {
        "interface_confidence": 91.2,
        "positional_accuracy": 81.0,
        "contact_extent": 90.0,
        "length_suitability": 100.0
      },
      "raw_metrics": {
        "interface_plddt_raw": 91.2,
        "pae_interchain_raw": 3.8,
        "contact_count_raw": 18,
        "peptide_length": 10
      }
    }
  ],
  "priorities_for_validation": ["PEP_042", "PEP_017", "PEP_089", "PEP_023", "PEP_056"]
}
```

## Error Handling

| Error | Detection | Handling |
|-------|-----------|----------|
| Invalid amino acids | parse_inputs.py | Skip peptide, log to stderr |
| Target JSON format error | parse_inputs.py | Stop with error message |
| Prediction timeout | AlphaFold Server/Local | Retry once, then mark as failed |
| No structure file found | extract_metrics.py | Mark as failed, continue |
| PAE matrix missing | extract_metrics.py | Use default value (99) |
| No predicted contacts | calculate_contacts | Assign 0 contacts |
| Empty rankings | Final validation | Warn user, suggest checking predictions |

## Success Criteria

The pipeline is considered successful when:

1. **All peptides processed**: No crashes or unhandled exceptions
2. **Metrics extracted**: Each prediction yields pLDDT, PAE, contact count
3. **Clear ranking**: Sorted list with score distribution across categories
4. **Interpretable report**: Human-readable with methodology documented
5. **Limitations stated**: Explicit acknowledgment of computational limitations

## Limitations and Scientific Caveats

### Computational Limitations

1. **AlphaFold 3 Accuracy**
   - Predictions are computational hypotheses
   - May fail for intrinsically disordered peptides
   - Membrane proteins remain challenging
   - Very short peptides (< 5 aa) often poorly predicted

2. **Static Structure Assumption**
   - AlphaFold 3 predicts a single conformation
   - Does not capture conformational ensembles
   - Missing dynamic behavior upon binding

3. **Missing Biological Context**
   - No post-translational modifications
   - No cellular concentrations
   - No competitive binders
   - No allosteric effects

### Scientific Limitations

1. **Binding ??Activity**: A peptide may bind without producing desired biological effect
2. **Specificity not guaranteed**: High-scoring peptides may bind multiple targets
3. **Affinity not quantified**: Scores indicate binding likelihood, not binding strength (Kd)
4. **Cellular context matters**: In vitro binding may not translate to cellular activity

### Mitigation Recommendations

1. Always validate predictions experimentally
2. Consider multiple scoring criteria beyond composite score
3. Test peptide variants (alanine scanning) for key residue identification
4. Account for peptide stability in therapeutic contexts
5. Use orthogonal methods for validation

## Benchmark Datasets

For validation and calibration:

- **PepBDB**: Protein-Peptide Binding Database (curated complexes)
- **SKEMPI 2.0**: Database of binding affinity changes upon mutation
- **PDB**: Filter for peptide-protein complexes (biological assemblies)

## References

1. Abramson et al. "Accurate structure prediction of biomolecular interactions with AlphaFold 3." Nature, 2024.

2. Ternavor et al. "A deep-learning framework for multi-level peptide-protein interaction prediction." Nature Communications, 2021.

3. Peterson et al. "PepCNN deep learning tool for predicting peptide binding residues." Scientific Reports, 2023.

4. Jumper et al. "Highly accurate protein structure prediction with AlphaFold." Nature, 2021.

5. De Vries et al. "Modeling complexes of peptides and proteins with HADDOCK." Structure, 2020.