Skip to Content
Introduction

HEDGEHOG

Hierarchical Evaluation of Drug GEnerators tHrOugh riGorous filtration

HEDGEHOG is a comprehensive benchmark pipeline for evaluating generative models in molecular design. It takes a set of generated SMILES strings, unify them during preprocessing, and progressively filters through physicochemical descriptors, structural alerts, synthetic accessibility checks, molecular docking, and post-docking pose analysis. It is suggested to run Boltz-2 for more precise binding affinity analysis alongside with docking evaluation. HEDGEHOG produces a detailed HTML report with per stage statistics, interactive visualizations, and generative quality metrics.

Pipeline Structure

Key Features

  • 6-stage evaluation pipeline: molecules pass through (i) preprocessing, (ii) descriptors, (iii) structural filters, (iv) synthesis evaluation, (v) docking, and (vi) post-docking three-dimensional (3D) filters.
  • 28 physicochemical dscriptors: size- and composition-related descriptors, lipophilicity and polarity descriptors, hydrogen bond features, ring statistics, fraction of sp3-hybridized carbon atoms, and additional medicinal chemistry checks.
  • 11 structural filters: 2,458 SMART patterns for common alerts, including PAINS, Glaxo, BMS, Dundee filters, and others related, molecular graph severity, molecular complexity, NIBR, Bredt rule, Lilly MedChem rules, and five medicinal chemistry check, including protecting groups, ring infraction, stereo center, halogenicity, and symmetry.
  • Retrosynthesis analysis: synthetic accessibility evaluation with SA, RA, SYBA, and others related scores, and retrosynthetic route analysis via AiZynthFinder.
  • Molecular docking: binding affinity scoring and pose generation with SMINA, GNINA, and Matcha.
  • 3D post-docung checks: search-box containment, pose quality assessment, protein-ligand interaction checks, and a conformer-deviation analysis.
  • Automated reporting: interactive HTML report with molecule retention funnel, descriptor distributions (with reference threshold lines), filter breakdowns, docking results, and MolEval generative metrics.
  • CLI and TUI interfaces: a Typer-based command-line interface (hedgehog / hedge) for scripted workflows, plus an interactive terminal UI built with Node.js for visual configuration and pipeline management.

Quick Example

git clone https://github.com/LigandPro/hedgehog.git
cd hedgehog
uv sync
 
# First safe run: descriptors + structural filters only
uv run hedgehog --stage descriptors --stage struct_filters --force-new
Last updated on
Getting Started