---
name: alphafold3-mutation-impact-protocol
description: Predict how point mutations affect protein structure by comparing wild-type and mutant AlphaFold 3 predictions, assessing stability and interface changes.
allowed-tools: WebFetch, Bash(python *), Bash(mkdir *), Bash(cp *), Bash(ls *), Bash(jq *), Bash(cd *)
---

# AlphaFold 3 Mutation Impact Analyzer Protocol

## Purpose

Assess the structural impact of point mutations by comparing AlphaFold 3 predictions of wild-type and mutant protein structures. Generate quantitative metrics for stability changes, interface alterations, and functional implications.

## Inputs

Create an `inputs/` directory containing:

- `inputs/wildtype.json`: AlphaFold 3 JSON for the wild-type protein.
- `inputs/mutations.tsv`: Tab-separated file of mutations to analyze.
  ```
  mutation_id	protein_chain	original_residue	position	new_residue	notes
  MUT001	A	G	42	V	ClinVar likely pathogenic
  MUT002	A	E	105	K	 cancer hotspot
  ```
- `inputs/metadata.md`: Protein name, function, known domains, active site residues, known binding interfaces.

## Pre-Run Checks

1. Confirm research use is permitted.
2. Validate wild-type sequence uses standard amino acid codes.
3. Verify all mutations are valid (original residue matches sequence at position).
4. Check that position numbers are 1-indexed (convert if 0-indexed in file).
5. Ensure mutation is not at already-low-confidence region in wild-type.

## Step 1: Wild-Type Prediction

Run AlphaFold 3 prediction for the wild-type structure:

### Route A: AlphaFold Server

Submit wild-type job and download to `outputs/wildtype/`.

### Route B: Local

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

Store the wild-type structure and confidence files.

## Step 2: Generate Mutant Sequences

For each mutation in `inputs/mutations.tsv`:

1. Extract the protein sequence from the input JSON.
2. Verify position and original residue match.
3. Replace the residue at that position.
4. Create mutant JSON with updated sequence.
5. Store as `inputs/mutants/<mutation_id>.json`.

Example Python script:
```python
import json

with open('inputs/mutations.tsv') as f:
    header = f.readline()
    for line in f:
        parts = line.strip().split('\t')
        mut_id, chain, orig, pos, new = parts[0], parts[1], parts[2], int(parts[3])-1, parts[4]
        # Load and modify sequence
        seq = sequences[chain]
        new_seq = seq[:pos] + new + seq[pos+1:]
        # Save mutant JSON
```

## Step 3: Mutant Predictions

For each mutant:

### Route A: AlphaFold Server

1. Create new job with mutant sequence.
2. Submit and download to `outputs/mutants/<mutation_id>/`.

### Route B: Local

```bash
python run_alphafold.py \
  --json_path=inputs/mutants/<mutation_id>.json \
  --output_dir=outputs/mutants/<mutation_id>
```

## Step 4: Compare Structures

For each mutation pair (wild-type vs mutant):

Calculate comparison metrics:

1. **Overall RMSD** (excluding flexible ends)
2. **Local RMSD** around mutation site (± 10 residues)
3. **pLDDT difference** at mutation site
4. **PAE change** at known interface positions
5. **Side-chain volume change** (simple: V→L larger, E→K charge reversal)

Generate `outputs/comparison/<mutation_id>_comparison.json`:

```json
{
  "mutation_id": "MUT001",
  "mutation": "G42V",
  "chain": "A",
  "overall_rmsd": 1.2,
  "local_rmsd_10A": 2.8,
  "wildtype_plddt_at_site": 85.3,
  "mutant_plddt_at_site": 72.1,
  "plddt_change": -13.2,
  "predicted_impact": "significant",
  "impact_explanation": "Large local RMSD and pLDDT drop suggest structural disruption"
}
```

## Step 5: Categorize Impact

Classify each mutation:

| pLDDT change | Local RMSD | Impact Category |
|-------------|------------|-----------------|
| > -10 | > 2.0 | Severe |
| -5 to -10 | 1.0-2.0 | Moderate |
| -3 to -5 | 0.5-1.0 | Mild |
| > -3 | < 0.5 | Negligible |

## Step 6: Generate Report

Write `outputs/mutation_analysis.md`:

```markdown
# Mutation Impact Analysis Report

## Protein
- Name: [protein_name]
- Uniprot/Source: [ID]
- Length: [N] residues
- Known domains: [list]
- Active site residues: [positions]

## Methodology
- Prediction tool: AlphaFold 3
- Comparison: pairwise structure alignment (wild-type vs mutant)
- Impact criteria: [table above]

## Results Summary
- Total mutations analyzed: [N]
- Severe: [N]
- Moderate: [N]
- Mild: [N]
- Negligible: [N]

## Detailed Results

### [Mutation ID]: [mutation_string]
- Location: Chain [X], residue [N]
- Category: [Severe/Moderate/Mild/Negligible]
- pLDDT change: [value]
- Local RMSD: [value] nm
- Explanation: [interpretation]
- Correlation with clinical notes: [if provided]

## Pathogenicity Predictions
Based on structural impact:
- Likely pathogenic (structural disruption): [list]
- Uncertain (moderate changes): [list]
- Likely benign (minimal change): [list]

## Limitations
- AlphaFold 3 predictions are computational hypotheses
- Does not account for:
  - Protein dynamics and folding kinetics
  - Post-translational modifications
  - Protein-protein interaction effects beyond local structure
  - Functional sites distant from the mutation
- Severe structural change does not prove pathogenicity
- Conservative mutations can be pathogenic through mechanism not captured here

## Recommendations
1. Validate severe-impact mutations with experimental assays (thermal stability, binding assays)
2. Consider clinical variant databases (ClinVar, COSMIC) for validation
3. Run molecular dynamics for mutations near functional sites
4. Test protein-protein interactions if mutation is at interface

## References
- AlphaFold 3: Abramson et al., Nature, 2024
- Variant effect prediction: https://varify.bio
```

## Success Criteria

- Wild-type prediction completes successfully.
- All mutations are correctly applied without sequence errors.
- Comparison metrics are computed for each mutation.
- Impact categorization is consistent and documented.
- Report captures both quantitative metrics and biological interpretation.
- Limitations acknowledge computational nature of predictions.

## Failure Modes

- Invalid mutation (wrong residue at position) → skip, log error
- Mutation at low-confidence region in wild-type → note limitation
- Prediction fails for mutant → retry, if persistent, mark as "prediction failed"
- Identical structures → check if mutation is synonymous (amino acid same)

## References

- Richards et al., A database of clinically relevant variants, Hum Mut, 2018 (ClinVar)
- Lek et al., Analysis of protein-coding genetic variation, Nature, 2016 (gnomAD)
- AlphaFold 3: Abramson et al., Nature, 2024