---
name: alphafold3-cross-species-protocol
description: Predict and compare protein structures across multiple species to identify conserved structural elements and evolutionary relationships using AlphaFold 3.
allowed-tools: WebFetch, Bash(python *), Bash(mkdir *), Bash(cp *), Bash(ls *), Bash(jq *), Bash(cd *)
---

# AlphaFold 3 Cross-Species Comparative Structurome Protocol

## Purpose

Predict protein structures across multiple species and analyze conservation of structural elements, enabling evolutionary analysis and identification of conserved functional regions. This workflow supports comparative genomics and phylogenetics research.

## Inputs

Create an `inputs/` directory containing:

- `inputs/orthologs.fasta`: Multiple sequence alignment (MSA) or collection of ortholog sequences.
  ```
  >Human
  MVWALLVLLAALAG...
  >Mouse
  MVWALLAVLALAG...
  >Zebrafish
  MVWALLAVLALAG...
  >Drosophila
  MAWALLAVLVLAG...
  ```
- `inputs/species_list.tsv`: Tab-separated species information.
  ```
  species	common_name	divergence_time	annotation
  Homo_sapiens	Human	0	reference
  Mus_musculus	Mouse	90	well-annotated
  Danio_rerio	Zebrafish	450	fish model
  Drosophila_melanogaster	Fruit fly	720	invertebrate
  ```
- `inputs/metadata.md`:
  - Protein family name
  - Known domain architecture
  - Key functional residues
  - Reference structure (if available in PDB)

## Pre-Run Checks

1. Confirm research use is permitted.
2. Validate all sequences use standard amino acid codes.
3. Verify sequence alignment is reasonable (no large indels causing misalignment).
4. Check for gene duplicates or splice variants - include the main isoform.
5. Note highly divergent sequences may not align well.

## Step 1: Prepare Individual Species Inputs

For each species, create individual AF3 inputs:

```json
{
  "name": "protein_Homo_sapiens",
  "sequences": [
    {
      "protein_chain": {
        "sequence": "MVWALLVLLAALAG...",
        "id": {"value": "A"},
        "description": "Homo sapiens ortholog"
      }
    }
  ]
}
```

Organize as:
```
inputs/species/
  homo_sapiens.json
  mus_musculus.json
  danio_rerio.json
  drosophila_melanogaster.json
```

## Step 2: Predict Structures for Each Species

For each species:

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

**For AlphaFold Server**: Submit one job per species.

## Step 3: Generate Comparative Metrics

For each predicted structure:

```json
{
  "species": "Homo sapiens",
  "common_name": "Human",
  "pLDDT_mean": 89.2,
  "pLDDT_by_region": {
    "N-terminal (1-100)": 92.1,
    "Core domain (101-300)": 91.5,
    "C-terminal (301-400)": 78.3
  },
  "structured_regions": [10, 11, 12, 50, 51, 52],
  "disordered_regions": [1, 2, 3, 320, 321, 322],
  "predicted_features": ["alpha-helix", "beta-sheet"],
  "assembly_state": "homodimer"
}
```

## Step 4: Structure Alignment and Comparison

Perform pairwise structural alignments:

```python
# Using TM-align or similar
# Align each species structure to reference (e.g., Human)
```

Generate comparison matrix:

```json
{
  "reference_species": "Homo sapiens",
  "tm_scores": {
    "homo_sapiens": 1.0,
    "mus_musculus": 0.97,
    "danio_rerio": 0.85,
    "drosophila_melanogaster": 0.72
  },
  "rmsds_to_reference": {
    "homo_sapiens": 0.0,
    "mus_musculus": 1.2,
    "danio_rerio": 3.8,
    "drosophila_melanogaster": 6.5
  },
  "conserved_regions": [
    {"start": 50, "end": 150, "tm_score_range": [0.95, 1.0], "annotation": "catalytic core"},
    {"start": 200, "end": 280, "tm_score_range": [0.88, 1.0], "annotation": "binding interface"}
  ],
  "divergent_regions": [
    {"start": 1, "end": 30, "divergence": "high", "annotation": "N-terminal extension"},
    {"start": 300, "end": 350, "divergence": "moderate", "annotation": "species-specific insertion"}
  ]
}
```

## Step 5: Conservation Analysis

Map evolutionary conservation onto structure:

1. **Sequence conservation**: Calculate via alignment
2. **Structural conservation**: Compare Cα positions across species
3. **Functional residue conservation**: Check known active site residues

```json
{
  "conservation_analysis": {
    "overall_sequence_identity_range": "35-100%",
    "core_secondary_structure": "highly conserved",
    "active_site_residues": {
      "H98": {"conservation": "100%", "function": "catalytic"},
      "E200": {"conservation": "95%", "function": "substrate binding"},
      "H220": {"conservation": "100%", "function": "catalytic"}
    },
    "structurally_conserved": ["alpha-helix 1", "beta-sheet core"],
    "structurally_divergent": ["N-terminal arm", "C-terminal tail", "loop regions"]
  }
}
```

## Step 6: Generate Comparative Report

Write `outputs/cross_species_analysis.md`:

```markdown
# Cross-Species Comparative Structurome Analysis

## Protein Family
- Family name: [name]
- Pfam domain: [ID]
- Function: [description]

## Species Analyzed
| Species | Common Name | Sequence Length | Divergence (MYA) |
|---------|-------------|-----------------|------------------|
| Homo sapiens | Human | [N] | 0 |
| Mus musculus | Mouse | [N] | 90 |
| Danio rerio | Zebrafish | [N] | 450 |

## Prediction Quality by Species

| Species | Mean pLDDT | Confidence | Notes |
|---------|------------|------------|-------|
| Human | [N] | High | reference |
| Mouse | [N] | High | - |
| Zebrafish | [N] | Medium | - |

## Structural Comparison

### Global Similarity
| Comparison | TM-Score | RMSD (Å) | Assessment |
|------------|----------|----------|------------|
| Human vs Mouse | [N] | [N] | Highly similar |
| Human vs Zebrafish | [N] | [N] | Similar core |
| Human vs Fly | [N] | [N] | Core conserved |

### Structural Alignment
- Core domain alignment: [quality assessment]
- Invariant regions: [list with positions]
- Variable insertions: [list with species specificity]

## Conservation Analysis

### Sequence Conservation
- Mean pairwise identity: [N]%
- Most conserved region: residues [N-N], [annotation]
- Most variable region: residues [N-N], [annotation]

### Structural Conservation
| Region | Structural Variation | Functional Implication |
|--------|--------------------|----------------------|
| [N-N] | Low (TM > 0.9) | Conserved fold |
| [N-N] | Moderate | Adaptive region |

### Functional Site Conservation
| Residue | Position | Conservation | Function |
|---------|----------|--------------|----------|
| [H98] | [N] | [N]% | [catalytic/etc.] |
| [E200] | [N] | [N]% | [binding/etc.] |

## Evolutionary Insights

### Core Structure Evolution
- Core fold established by: [evolutionary age]
- Last universal common ancestor likely had: [description]

### Adaptive Evolution
- Species-specific insertions: [list with species]
- Functional diversification: [if observed]

### Phylogenetic Signal
- Structural data supports: [phylogenetic relationship]
- Conflicts with sequence-based tree: [yes/no, explanation]

## Species-Specific Features

### [Species name]
- Unique insertions: [list]
- N/C-terminal extensions: [description]
- Functional implications: [if known]

## Limitations
- AlphaFold 3 predictions are computational hypotheses
- Predicted structures may differ from actual experimental structures
- Very fast-evolving proteins may not align well
- Horizontal gene transfer may confuse orthology assignment
- Does not account for:
  - Post-translational modifications differences
  - Expression level variations
  - Subcellular localization changes
  - Neofunctionalization events

## Recommendations
1. Validate with experimental structures where available
2. Test functional differences experimentally
3. Analyze expression patterns across species
4. Consider structural phylogenetics alongside sequence phylogenetics
5. Investigate species-specific insertions for novel functions

## References
- AlphaFold 3: Abramson et al., Nature, 2024
- TM-align: Zhang & Skolnick, Nuc Acid Res, 2005
- Evolutionary conservation: Panchenko et al., Nat Methods, 2004
- Orthology: Altenhoff & Dessimoz, Trends Biochem Sci, 2009
```

## Success Criteria

- Structures are predicted for all species.
- Structural comparisons are quantified (TM-score, RMSD).
- Conservation patterns are mapped.
- Evolutionary insights are derived.
- Limitations acknowledge prediction limitations.

## Failure Modes

- Highly divergent sequences fail to align → may need to predict domains separately
- Very low TM-scores → protein may have different folds (consider if truly orthologs)
- Missing key species → add intermediate species for better phylogeny

## References

- AlphaFold 3: Abramson et al., Nature, 2024
- TM-align: Zhang & Skolnick, Nuc Acid Res, 2005
- Protein evolution: Koonin, Annu Rev Genet, 2005
- Orthology inference: Altenhoff & Dessimoz, Trends Biochem Sci, 2009