The first AI drug discovery game for students. Explore real 3D protein structures, battle pathogens, and generate candidate molecules — powered by Google Gemma running locally on your machine.
GemmaCure bundles everything — Python, Flask, and all dependencies — inside the Mac app. All you need is an Apple Silicon Mac and a Gemma endpoint.
Download the .dmg and drag to Applications. Python, Flask, RDKit, and the entire backend are bundled — nothing to install separately.
A running Gemma model server reachable over your network (e.g. on a DGX workstation or local GPU). The app connects on first launch.
No Python installation. No package managers. No command line. Open the app, enter your LLM endpoint URL, and you're doing drug discovery in under 60 seconds.
Unlike apps that just show animations, GemmaCure uses real biochemistry data, real SMILES molecule validation, and a real LLM to generate drug candidates.
Explore real protein structures from the RCSB Protein Data Bank and AlphaFold. Rotate, zoom, and inspect the binding site your molecule will target. Every target is a real disease-relevant protein.
Gemma generates candidate drug molecules as SMILES strings, validated in real-time with RDKit. See QED score, synthetic accessibility, LogP, and Lipinski's Rule of Five — the same metrics real medicinal chemists use.
Battle 20 real pathogens — from influenza to MRSA — in a turn-based format. Each round you design a molecule to fight the target. Win XP, unlock achievements, level up.
Free-form drug design mode. Pick any target, load its structure, run the LLM multiple times, and compare candidate molecules side-by-side in the discovery popup.
Every generated molecule is explained in plain English: what QED means, why LogP matters, whether the molecule "looks like a drug." Science education, not just science output.
Level up as you discover better molecules. Earn badges for milestones, track your experiment history, and share discoveries with classmates.
Download the .dmg, drag GemmaCure to your Applications folder, and open it. Python 3.13, Flask, RDKit, and all dependencies are bundled inside — nothing else to install.
On first launch, enter your Gemma LLM endpoint URL (e.g. http://dgx1:9002). The app sends a test drug design prompt and validates the connection before saving.
Pick from 20 curated disease targets — cancer, viruses, bacteria, and more. Load its 3D protein structure from RCSB or AlphaFold, see the binding site you need to block.
Run the AI to generate candidate molecules. Explore each one in the discovery popup — 3D protein view, 2D molecule drawing, drug property scores, and plain-English explanations of what it all means.
| Textbooks | Web simulators | GemmaCure | |
|---|---|---|---|
| Real protein structures | ❌ | ⚠️ Static images | ✅ Live RCSB + AlphaFold |
| AI molecule generation | ❌ | ❌ | ✅ Google Gemma LLM |
| Drug-likeness validation | ❌ | ❌ | ✅ RDKit, QED, Lipinski |
| Works offline | ✅ | ❌ | ✅ Fully local |
| Gamified learning | ❌ | ⚠️ Limited | ✅ XP, battles, badges |
| Plain-English explanations | ⚠️ Dense | ⚠️ Minimal | ✅ Student-friendly |
| Cost | 💰 Expensive | 💰 Subscription | ✅ Free forever |
| Data privacy | N/A | ❌ Cloud-dependent | ✅ 100% on-device |
Every target is a real protein involved in human disease, sourced from peer-reviewed research and the RCSB Protein Data Bank.
Everything runs locally on your machine. Your molecule designs, experiment history, and LLM queries never leave your network.
The Flask backend, SQLite database, and all experiment data live entirely on your Mac. No cloud account required.
We never see your molecule designs, protein targets, or LLM prompts. The app has no telemetry, no analytics, no tracking.
GemmaCure connects only to the endpoint URL you provide — typically a local GPU server on your school or lab network.
GemmaCure is designed exclusively for educational purposes to inspire curiosity about drug discovery and the potential of artificial intelligence in medicine. We believe AI has the potential to transform how we approach disease, and this app tries to bring that excitement to students and young learners.
Nothing in this application constitutes medical advice, diagnosis, or treatment. Never use any output from GemmaCure to make health or medical decisions. Always consult a qualified medical professional.
Molecules generated by this app are AI-assisted simulations for learning purposes only. They have not been synthesised, tested, validated, or approved by any regulatory authority. Real drug discovery requires years of rigorous laboratory and clinical research.
The Gemma language model may produce inaccurate, incomplete, or nonsensical molecule structures. Drug-likeness scores are computational estimates and do not predict real-world biological activity, safety, or efficacy.
GemmaCure is provided "as is" without warranty of any kind, express or implied, including but not limited to warranties of merchantability, fitness for a particular purpose, or non-infringement. Use at your own risk.
Protein structures are sourced from RCSB PDB and AlphaFold DB for educational purposes. These databases are subject to their own terms of use. GemmaCure does not claim ownership of any biological data.
GemmaCure is designed for students aged 13 and above under appropriate supervision. It is a tool for sparking interest in STEM — not a professional research platform.
Free, open source, and ready to run. Download the Mac app or clone the repo to get started in minutes.
xattr -cr "/Applications/GemmaCure.app" in Terminal.
Whether you're a student, educator, or researcher — we'd love to hear from you. Report bugs, suggest new targets, or ask about using GemmaCure in your classroom.