🧪 Qrucible — Quantum‑Validated Molecular Screening Platform

Qrucible is a real‑world–aligned, decision‑support system for early‑stage drug discovery.
It combines classical machine learning (QSAR/SAR) with physics‑grounded quantum feasibility checks to reduce false positives before wet‑lab synthesis.

Core idea:
Use ML to screen fast, then use quantum methods as a mandatory feasibility gate — not as a novelty.

---

🔍 Why Qrucible Exists:

In real pharmaceutical pipelines:

• ML/QSAR models generate many false positives

• Chemically implausible molecules survive virtual screening

• Wet‑lab synthesis and assays are expensive and slow

• Physics‑based validation is often skipped due to cost

Qrucible solves this gap by inserting a quantum feasibility gate between ML screening and wet‑lab experiments.

---

🎯 What Qrucible Does:

• Accepts researcher‑provided molecular libraries

• Performs classical ML screening for target properties

• Applies medicinal chemistry rule filters

• Uses quantum algorithms to reject physically implausible candidates

• Outputs a short, synthesis‑ready molecule list

---

🧠 High‑Level Pipeline (Version 4)

Researcher Molecules
        ↓
Classical ML Screening (QSAR / SAR)
        ↓
Chemical Feasibility Filters
        ↓
Quantum Feasibility Gate (YES / NO)
        ↓
Shortlist for Wet‑Lab Synthesis

---

🧬 Detailed Workflow

STEP 0 — Researcher Input

• Existing molecules (hits, analogs, fragments)

• Target property objectives (e.g., pIC50)

• Optional target context

📌 Matches real pharma workflows

---

STEP 1 — Classical ML Screening

• Molecular descriptors + fingerprints (RDKit)

• Models: Random Forest, XGBoost (optionally GNNs)

• Outputs:

  • Predicted activity

  • Prediction uncertainty

  • Ranking

📌 High‑throughput, noisy by design — just like industry ML

---

STEP 2 — Chemical Feasibility Filters

• Lipinski / Veber rules

• Toxic or reactive substructures

• Extreme phys‑chem removal

• Simple synthetic accessibility heuristics

📌 Medicinal chemistry sanity check

---

STEP 3 — Quantum Feasibility Gate (Core Contribution)

Quantum methods are used only for decision‑critical validation, not ranking.

Quantum checks include:

• Electronic stability

• Ground‑state energy consistency

• Conformational feasibility

• Detection of unphysical energy landscapes

Decision logic:

• ❌ Reject if physically implausible

• ✅ Pass if feasible

📌 This step is non‑removable and justifies quantum usage

---

STEP 4 — Final Output

• 10–30 shortlisted molecules

• For each candidate:

  • ML score + uncertainty

  • Quantum pass/fail

  • Rejection rationale (if any)

📌 Direct input to wet‑lab synthesis

---

⚛️ Role of Quantum Computing:

• Physics‑based feasibility validation

• False‑positive rejection

• Pre‑synthesis decision making

---

🧰 Tech Stack

Classical ML

• Python 3.10

• RDKit (conda‑forge)

• NumPy, Pandas

• Scikit‑learn

• XGBoost

• Joblib

Quantum

• Qiskit / PennyLane

• Quantum simulators (NISQ‑compatible)

• VQE / QAOA (small systems only)

Data

• ChEMBL (primary source)

• Curated target‑specific datasets

• Descriptor + fingerprint representations

---

🚀 Intended Use Case

Qrucible is designed for:

• Hit‑to‑lead optimization

• SAR refinement

• Reducing wet‑lab failures

• Academic + industry‑aligned research

It is not intended as a fully autonomous drug discovery system.

---

Project Structure

qrucible/
├── src/              # Source code
├── notebooks/        # Jupyter notebooks
├── data/            # Datasets (raw, processed, interim)
├── results/         # Models, figures, reports
├── scripts/         # Executable scripts
├── tests/           # Unit tests
└── config/          # Configuration files

Quick Start

1. Activate environment:

   conda activate qrucible

2. Download data:

   python scripts/download_data.py

3. Preprocess:

   python src/data/preprocessor.py

4. Train models:

   python scripts/train_classical.py

5. Start Project:

   streamlit run app.py

Requirements

• See requirements.txt for full list

---

📈 Results

Data Statistics

• Source: ChEMBL EGFR Kinase inhibitors
• Downloaded: 16,449 compounds
• Cleaned: 9,224 compounds
• Activity Range: pIC50 5.00 - 17.30
• Mean Activity: pIC50 7.20 ± 1.22

Model Files

• results/models/random_forest.pkl
• results/models/xgboost.pkl
• results/models/random_forest_fingerprints.pkl
• results/models/xgboost_fingerprints.pkl

Visualizations

• Prediction plots: results/figures/
• Model comparison: results/figures/model_comparison.png

---

Status: Version 3 completed ✅ | Version 4 Phase 1 In Progress 🔄

%% Last Updated: 2025-10-12 %%

%% Performance: Test R² = 0.64 | Pearson r = 0.804 %%

---

🤝 Contributing

This is a research project. For questions or collaboration:

• Open an issue
• Submit a pull request
• Contact: vdixit2505@gmail.com | rajhansdhruv@gmail.com

📄 License

MIT License

🙏 Acknowledgments

• ChEMBL Database for molecular data
• RDKit for cheminformatics toolkit
• IBM Qiskit for quantum computing framework

📞 Support

For issues or questions:

1. Check documentation in docs/
2. Review GETTING_STARTED.md
3. Open an issue on GitHub

---