hedgehog 1.0.4

Hierarchical Evaluation of Drug GEnerators tHrOugh riGorous filtration

🦔 HEDGEHOG

Hierarchical Evaluation of Drug GEnerators tHrOugh riGorous filtration

Comprehensive benchmark pipeline for evaluating generative models in molecular design.

Pipeline Stages:

Each stage takes the output of the previous one, progressively filtering the molecule set:

1. Mol Prep (Datamol): salts/solvents & fragments cleanup, largest-fragment selection, metal disconnection, uncharging, tautomer canonicalization, stereochemistry removal → produces standardized “clean” molecules (molPrep folder)
2. Molecular Descriptors: 22 physicochemical descriptors (logP, HBD/HBA, TPSA, QED, etc.) → molecules outside thresholds are removed (descriptors folder)
3. Structural Filters: 6 criteria with ~2500 SMARTS patterns (PAINS, Glaxo, NIBR, Bredt, etc.) → flagged molecules are removed (structural filters folder)
4. Synthesis Evaluation: SA score, SYBA score, AiZynthFinder retrosynthesis → unsynthesizable molecules are removed (synthesis folder)
5. Molecular Docking: SMINA and/or GNINA → binding affinity scoring (docking folder)
6. Docking Filters: post-docking pose quality filtering → poor binders are removed
7. Final Descriptors: recalculation on the filtered set

Post-pipeline analysis: MolEval generative metrics

Setup & Run

Install from PyPI

python -m pip install hedgehog
hedgehog --help

Base install is intentionally lightweight and works on modern Python versions (including Python 3.13) without optional heavy docking extras.

Optional extras:

# Legacy PoseCheck backend for docking filters
python -m pip install 'hedgehog[docking-legacy]'

# Shepherd-Score Python dependency only (may be unavailable on some ABIs, e.g. cp313)
python -m pip install 'hedgehog[shepherd]'

Recommended Shepherd setup is an isolated worker environment:

uv run hedgehog setup shepherd-worker --yes

Install from source (recommended for development)

# Clone repository
git clone https://github.com/LigandPro/hedgehog.git
cd hedgehog


# Install AiZynthFinder (for synthesis stage) - recommended CLI flow
uv run hedgehog setup aizynthfinder --yes

# Legacy helper script (alternative)
./modules/install_aizynthfinder.sh

# Install package with uv
uv sync

You are ready to use 🦔 HEDGEHOG for your purpose!

Usage

# Run full pipeline on a proposed small test data from `data/test/`
uv run hedgehog run

# Alternatively, using the short-name alias:
uv run hedge run

# Run specific stage
uv run hedge run --stage descriptors

# Auto-install missing optional external tools during a run
uv run hedge run --auto-install

# Reuse the existing results folder
uv run hedge run --reuse

# Force a fresh results folder for stage reruns
uv run hedge run --stage docking --force-new

# Enable live progress bar in CLI
uv run hedge run --progress

# Regenerate HTML report from an existing run
uv run hedge report results/run_10

# Show pipeline stages and current version
uv run hedge info
uv run hedge version

# Get help
uv run hedge --help

Terminal UI (TUI)

For interactive configuration and pipeline management, use the TUI:

uv run hedgehog tui

If the TUI has not been built yet, the CLI will install/build it automatically on first launch. You can also launch it directly from the TUI package:

cd tui
npm run tui

See tui/README.md for details and developer workflow.

Unified verification pipeline

Use one command entry point for local/CI checks:

# Quick local smoke (CLI + TUI build + TUI startup/quit in PTY)
uv run python scripts/check_pipeline.py --mode quick

# CI smoke profile (same checks, no full production run)
uv run python scripts/check_pipeline.py --mode ci

# Full local verification (quick checks + full production pipeline run)
uv run python scripts/check_pipeline.py --mode full

--mode full runs uv run hedgehog run with the default production config, so it can be long-running and requires external stage dependencies (for example docking/synthesis tooling) to be installed in your local environment.

Git hooks with Lefthook (recommended)

Use Lefthook to block commits/pushes that would fail CI:

# Install Lefthook (macOS)
brew install lefthook

# Register git hooks from lefthook.yml
lefthook install

# Optional: run hooks manually
lefthook run pre-commit
lefthook run pre-push

Current local gates:

• pre-commit: staged Python formatting/lint (ruff) and whitespace checks.
• pre-push: repository-wide ruff checks, pytest, pipeline smoke (scripts/check_pipeline.py --mode ci), and docs build (docs && pnpm build).

If you need to skip a specific hook command once (not recommended), use SKIP:

SKIP=docs-build git push

Documentation Site

cd docs && pnpm install && pnpm dev

The docs site is built with Nextra and available at http://localhost:3000.

HTML Reports

After each pipeline run, an interactive HTML report is automatically generated as report.html in the results folder. The report includes:

• Pipeline summary and molecule retention funnel
• Per-stage statistics and visualizations
• Descriptor distributions
• Filter pass/fail breakdowns
• Synthesis scores and docking results

Configure your run Edit config for each stage in configs folder based on metrics you want to calculate.