molclaw-equiscore-tool

Files (1)

• SKILL.md

SKILL.md (first lines)

# EquiScore Multi-Tool Workflow

## Invocation

Use the exact tool name and parameter schema exposed by the current harness. Tool names in SDK examples below are raw MCP names, not model-facing aliases. Local files must be uploaded with `molclaw-file-transfer` before a server tool can use them. Repair PDB inputs first when the workflow requires it.

> [!NOTE]
> Local files are not directly accessible by the server. Please upload them to the server using `molclaw-file-transfer` before execution.
> For PDB file inputs, it is recommended to preprocess them using `molclaw-fix-pdb` before execution.

## Usage

### 1. Pocket Extraction
The description of tool *equiscore_pocket*.

```text
Extract binding pockets from docking results and prepare split single-molecule SDFs for EquiScore screening.
Args:
    docking_result (str): Path to a docking-result SDF file.
    receptor_pdb (str): Path to the receptor PDB file.
    pocket_cutoff (float|None): Optional numeric cutoff for pocket detection.
    dry_run (bool|None): If True, validate inputs and prepare outputs without executing EquiScore.
Return:
    status (str): 'success' or 'error'.
    msg (str): Human-readable summary or error message.
    command (str): The subcommand executed ('get_pocket').
    run_dir (str|None): Run-specific directory under tool_result/equiscore_result.
    single_sdf_dir (str|None): Path to directory containing split single-molecule SDFs.
    pocket_dir (str|None): Path to the generated pocket folder.
    split_sdf_count (int|None): Number of split SDF files created.
    pocket_item_count (int|None): Number of pocket entries generated.
    sample_single_sdfs (List[str]|None): Sample single-SDF filenames.
    sample_pockets (List[str]|None): Sample pocket directory names.
```

How to use tool *equiscore_pocket* :

```python
response = await client.session.call_tool(
    "equiscore_pocket",
    arguments={
        "docking_result": "relative/path/to/docking_result.sdf",
        "receptor_pdb": "relative/path/to/receptor.pdb",
        "pocket_cutoff": 10.0,
        "dry_run": True
    }
)
result = client.parse_result(response)
key_output = result["single_sdf_dir"]
```

#### Example parameter sets

```python
# 1) Main mode
{
    "docking_result": "relative/path/to/docking_result.sdf",
    "receptor_pdb": "relative/path/to/receptor.pdb",
    "pocket_cutoff": None,
    "dry_run": False
}

# 2) Variant mode
{
    "docking_result": "relative/path/to/docking_result.sdf",
    "receptor_pdb": "relative/path/to/receptor.pdb",
    "pocket_cutoff": 8.5,
    "dry_run": True
}
```

### 2. Pocket Screening
The description of tool *equiscore_screen*.

```text
Score a pocket library with EquiScore and return prediction CSV plus summary statistics.
Args:
    pocket_dir (str): Path to a pocket directory produced by `equiscore_pocket`.
    ngpu (int): Number of GPUs to use. Default: 1.
    batch_size (int): Inference batch size. Default: 128.
    num_workers (int): Number of worker processes for data loading. Default: 8.
    weight_path (str|None): Optional path to model weights.
    multi_pose (bool): If True, score multiple poses per ligand.
    pose_num (int): Number of poses to evaluate when `multi_pose` is True. Default: 1.
    debug (bool): Enable debug mode.
    dry_run (bool|None): If True, validate inputs without running EquiScore.
Return:
    status (str): 'success' or 'error'.
    msg (str): Human-readable summary or error message.
    command (str): The subcommand executed ('screen').
    run_dir (str|None): Run-specific directory under tool_result/equiscore_result.
    output_dir (str|None): Directory where screening outputs were written.
    predictions_path (str|None): Path to the CSV file with raw predictions.
    prediction_count (int|None): Number of prediction rows in the CSV.
    score_field (str|None): CSV column used for scoring, if detected.
    max_score (float|None): Maximum observed score.
    min_score (float|None): Minimum observed score.
    mean_score (float|None): Mean score.
    median_score (float|None): Median score.
```

#### Scoring Interpretation (EquiScore)

- EquiScore is trained as a classifier (`active=1`, `decoy=0`), so `0.5` can be used as a rough reference boundary.
- In practical virtual screening, absolute thresholding is less robust than ranking.
- Recommended usage:
    - Sort predictions by score column (commonly `test_pred`) in descending order.
    - Select top N or top N% compounds for downstream validation.
    - Typical settings include top 1% for enrichment-style filtering or top 50-200 molecules for follow-up.
- For higher confidence, combine EquiScore ranking with another docking/scoring method for consensus prioritization.

How to use tool *equiscore_screen* :

```python
response = await client.session.call_tool(
    "equiscore_screen",
    arguments={
        "pocket_dir": "relative/path/to/pockets",
        "ngpu": 1,

From InternScience/MolClaw, MIT.

More from InternScience/MolClaw