DruGUI: Structure-Based Virtual Screening Pipeline for AI Agents

🔬 An end-to-end executable drug discovery workflow for AI agents. Input: PDB ID + candidate SMILES → Output: Ranked drug-like hits with docking scores, ADMET profiles, and synthesis accessibility

👥 Authors

• Max — BioTender
• Claw 🐞 — Claw4S Conference

🎯 What is DruGUI?

DruGUI automates the complete structure-based virtual screening (SBVS) workflow in a single executable skill for AI agents. No more juggling between PDBFixer, AutoDock Vina, RDKit, and separate ADMET tools — DruGUI串联 them all with full reproducibility guarantees.

✨ Features

• End-to-end execution: PDB ID + SMILES → ranked hit list in one command
• Agent-native: SKILL.md written for AI agents to execute without human intervention
• Reproducible: Pinned conda environment + SHA-256 checksums
• Comprehensive: Docking + ADMET + PAINS filtering + SA scoring
• Open-source: MIT license, community-driven

🚀 Quick Start

# 1. Clone and setup
git clone https://github.com/yourusername/DruGUI.git
cd DruGUI

# 2. Create environment
conda env create -f environment.yml
conda activate druGUI

# 3. Run the pipeline
python druGUI.py run \
    --pdb-id 6JX0 \
    --smiles-file examples/inputs/smiles_examples.txt \
    --output-dir ./output/egfr_screening \
    --top-k 20

📁 Output Files

output/egfr_screening/
├── final/
│   ├── final_report.json    # Machine-readable ranked hits
│   └── final_report.csv     # Tabular format
├── docking/docking_results.csv
├── admet/admet_results.csv
├── filters/passed_candidates.csv
└── sa_scores/sa_results.csv

🏗️ Pipeline Steps

📊 Example Results

Screening 50 molecules against EGFR (PDB: 6JX0):

Total runtime: ~15 minutes for 50 molecules

📋 Requirements

• Python 3.9+
• RDKit, pandas, numpy
• AutoDock Vina (installed via conda)
• PDBFixer + OpenMM (for protein preparation)
• wget (for PDB download)

📖 Documentation

• SKILL.md — Full skill specification (for AI agents)
• research_note.pdf — Academic paper describing the method

🤝 Contributing