---
name: alphafold3-rna-rbp-protocol
description: Predict RNA secondary structure and RNA-protein binding interfaces using AlphaFold 3, with analysis of RBP interaction motifs and binding affinity indicators.
allowed-tools: WebFetch, Bash(python *), Bash(mkdir *), Bash(cp *), Bash(ls *), Bash(jq *), Bash(cd *)
---

# AlphaFold 3 RNA Structure & RBP Binding Predictor Protocol

## Purpose

Predict RNA secondary and tertiary structure, and analyze RNA-binding protein (RBP) interaction interfaces. This workflow combines AlphaFold 3 predictions for both RNA structure and RNA-protein complexes, supporting research on non-coding RNA function and post-transcriptional regulation.

## Inputs

Create an `inputs/` directory containing:

- `inputs/rna.json` or `inputs/rna.fasta`: RNA sequence(s) to predict.
- `inputs/rbp.json` (optional): AlphaFold 3 JSON for an RNA-binding protein to test binding.
- `inputs/rna_annotations.md` (optional): Known motifs (RBP binding sites, riboswitches, miRNA targets).
- `inputs/metadata.md`:
  - RNA type (mRNA, lncRNA, miRNA, snRNA, rRNA, viral RNA)
  - Organism source
  - Known functions, localizations
  - RBP candidates if testing binding

## Pre-Run Checks

1. Confirm research use is permitted.
2. Validate RNA sequence uses only A, U, G, C (no modified nucleotides initially).
3. Check for potential pseudoknots (AF3 may struggle with these).
4. Verify sequence length is appropriate for AF3:
   - Short RNAs (< 500 nt): Full structure prediction
   - Long RNAs (> 500 nt): Consider fragment-based approach or focus on domains
5. If testing RBP binding: verify protein is a known or suspected RBP.

## Step 1: RNA Structure Prediction

### Route A: AlphaFold Server

1. Create new job with RNA sequences.
2. Submit and wait for completion.
3. Download results to `outputs/rna_structure/`.

### Route B: Local AlphaFold 3

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

## Step 2: Analyze RNA Structure

Extract structural features:

1. **pLDDT scores**: Higher pLDDT indicates more confident structure
2. **Base pairing**: Identify paired regions
3. **Single-stranded regions**: Often functional (RBP binding, miRNA pairing)

```json
{
  "rna_id": "lncRNA_X",
  "length": 1500,
  "sequence": "AUGC...",
  "overall_confidence": "medium",
  "pLDDT_mean": 72.3,
  "pLDDT_distribution": {
    "highly_confident (> 90)": [100, 150, 200],
    "confident (70-90)": [250, 300, 350, 400],
    "low_confidence (< 70)": [500, 550, 600]
  },
  "predicted_stems": [
    {"start": 100, "end": 200, "pLDDT": 88.5, "length": 100},
    {"start": 400, "end": 480, "pLDDT": 85.2, "length": 80}
  ],
  "predicted_loops": [
    {"position": 201, "length": 48, "pLDDT": 65.3},
    {"position": 481, "length": 19, "pLDDT": 78.2}
  ],
  "disordered_regions": [500, 501, 502, 650, 651]
}
```

## Step 3: Predict RBP Binding (if applicable)

If testing RNA-protein binding:

### Prepare Complex Input

```json
{
  "name": "lncRNA_X + RBP_Y complex",
  "sequences": [
    {
      "rna_chain": {
        "sequence": "AUGC...",
        "id": {"value": "A"},
        "description": "lncRNA"
      }
    },
    {
      "protein_chain": {
        "sequence": "MRGA...",
        "id": {"value": "B"},
        "description": "RBP"
      }
    }
  ]
}
```

### Run Prediction

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

## Step 4: Analyze RBP Binding Interface

Extract binding metrics:

```json
{
  "rna": "lncRNA_X",
  "rbp": "RBP_Y",
  "interface_residues_rna": [150, 151, 152, 200, 201],
  "interface_residues_rbp": [80, 81, 82, 120, 121, 122],
  "interface_pLDDT_rna": 75.3,
  "interface_pLDDT_rbp": 88.4,
  "interface_pLDDT_mean": 81.9,
  "binding_confidence": "medium",
  "motif_identified": "RGG box-like",
  "contact_count": 35
}
```

## Step 5: Motif Analysis

Identify known RNA-binding motifs in the protein:
- RRM (RNA Recognition Motif): RNP1 octamer (K/R-G-F/Y-G/A-F/Y-V/L/I-X-F/Y)
- KH domain: [V/I]-G-X-X-G
- RGG box: multiple RGG repeats
- ZnF (CCHC): C-X2-C-X4-H-X4-C

## Step 6: Generate Report

Write `outputs/rna_rbp_analysis.md`:

```markdown
# RNA Structure & RBP Binding Analysis Report

## RNA Target
- Name: [name]
- Type: [lncRNA/mRNA/miRNA/etc.]
- Length: [N] nucleotides
- Source organism: [organism]
- Overall pLDDT: [value]
- Confidence assessment: [High/Medium/Low]

## Structural Features

### Predicted Helical Regions
| Start | End | Length | Confidence |
|-------|-----|--------|------------|
| [N]   | [N] | [N]    | [pLDDT]    |

### Predicted Loops and Junctions
| Position | Length | Confidence |
|----------|--------|------------|
| [N]      | [N]    | [pLDDT]    |

### Disordered/ Flexible Regions
- Regions: [list positions]
- Note: Often functional for interactions

## Functional Predictions

### Predicted Binding Sites (for RBPs)
[If RBP complex was predicted]
- RBP name: [name]
- Interface confidence: [High/Medium/Low]
- Binding motif identified: [motif name]
- Interface residues on RNA: [list]

### Potential Regulatory Elements
- miRNA target sites: [if predicted]
- RBP binding motifs: [consensus sequences]
- Splicing regulatory elements: [if applicable]

## RBP Analysis (if included)

### Protein
- Name: [RBP name]
- Known domains: [RRM, KH, ZnF, etc.]
- Previous RBP annotations: [source databases]

### Binding Assessment
- Binding predicted: [Yes/No/Uncertain]
- Confidence: [High/Medium/Low]
- Interface quality: [description]

## Biological Interpretation

### Potential Functions
[Based on predicted structure]
- [Function 1]: supported by [evidence]
- [Function 2]: supported by [evidence]

### Novel Predictions
- Previously uncharacterized structure: [yes/no]
- Novel binding interface predicted: [yes/no]

## Limitations
- AlphaFold 3 RNA modeling may not capture:
  - Pseudoknots
  - Long-range tertiary interactions
  - RNA modifications effects
  - Dynamic conformational changes
- Modified nucleotides not supported
- Long RNA (> 2000 nt) predictions may be unreliable
- Binding affinity cannot be predicted from structure alone
- Low-confidence regions may still be functional

## Recommendations
1. Validate predicted structure with chemical probing (SHAPE-seq)
2. Test RBP binding with CLIP-seq or RNA EMSA
3. Compare with known structures in PDB for similar RNAs
4. For long RNAs, consider fragment-based prediction or domain analysis
5. Investigate functional implications of disordered regions

## References
- AlphaFold 3: Abramson et al., Nature, 2024
- RNA structure prediction: Sun et al., Nat Methods, 2019
- RBP databases: Ray et al., Nature, 2013 ( CLIPdb)
```

## Success Criteria

- RNA structure is predicted with interpretable confidence.
- Structural elements (helices, loops, disordered regions) are identified.
- If RBP included: binding interface is analyzed.
- Report provides testable hypotheses.
- Limitations acknowledge AF3 limitations for RNA.

## Failure Modes

- RNA prediction fails → check for invalid characters
- Very low pLDDT throughout → RNA may be highly flexible/unstructured
- No predicted binding → may be true negative, or prediction limitation
- Unexpected structure → validate against known domains/motifs

## References

- AlphaFold 3: Abramson et al., Nature, 2024
- RNA structure: Dawson & Pettitt, Nuc Acid Res, 2024
- RBP motifs: Lunde et al., Nat Rev Mol Cell Bio, 2007
- RNA-seq databases: RNAcentral Consortium, Nuc Acid Res, 2023