Powered by Gemma LLM DLYog Lab · 2026 Educational Use Only Founded on Deep2Lead research · arXiv:2108.05183

Learn drug discovery
by doing.

Start from atoms and finish with an AI-designed candidate you can defend. Explore real protein structures, generate molecules, weigh the tradeoffs a real medicinal chemist weighs, and learn where quantum computing may—and may not—help early drug discovery.

🆓 Free core curriculum 🔬 Real science 🎮 Learn by playing 🌐 Runs in your browser 🎓 Educational only
Deep2Lead · Discovery Arena
⚔️ Discovery Arena
⚗️ Experiment Studio
📂 My Experiments
🦠 EGFR Kinase
Design a Candidate
Molecule
CC1=CC=C(NC2=NC=CC(=N2)C3=CN=CC=C3)C=C1
QED 0.82 Lipinski ✓ SAS 3.1

Free to Start

Deep2Lead's core curriculum is free — the Science Journey, the Discovery Arena, and the Experiment Studio. Premium early access funds advanced curriculum and learner tools; selecting it today records interest without charging you.

🧬 Start the Science Journey No install · No card
Watch

See It In Action

Start with the science — how atoms become AI-designed drug candidates — then watch the platform put it to work against disease bosses with live AI inference and 3D protein visualization.

🧬 Start here · The science Atoms to AI Designed Drugs

The first-principles journey — atoms, proteins, targets, binding, and where AI and quantum computing actually help. The video companion to the Science Journey.

⚔️ Product demo Deep2Lead in the Discovery Arena

Generating real drug candidates against disease bosses — live AI inference, 3D protein visualization, and the full game loop end to end.

Hands-on Labs

Real Science.
Real AI. Real Tradeoffs.

Deep2Lead is not an animation. It uses real protein structures, real SMILES validation, and a real LLM to generate candidates you then have to defend on the evidence.

🧬

Science Journey

A first-principles course from atoms to AI-designed drugs — proteins, disease targets, SMILES, binding, ADMET, toxicity, and multi-objective tradeoffs — with a 19-question FAQ answering what students actually ask.

8 chapters · Clickable report card · FAQ
🤖

AI Molecule Generation

Gemma generates candidate molecules as SMILES strings, validated in real time with RDKit. See QED, synthetic accessibility, LogP, and Lipinski's Rule of Five — the same metrics real medicinal chemists use.

RDKit · SMILES · Drug-likeness scoring
⚔️

Discovery Arena

Battle 7 disease bosses in a turn-based format. Each round you design a molecule to beat the known drug's score. Earn XP, unlock achievements, climb the leaderboard.

7 bosses · XP & badges
⚗️

Experiment Studio

Free-form design mode. Pick any of 20 curated disease targets, load its structure, run the model repeatedly, and compare candidates side by side as a saved, reproducible experiment.

20 curated targets · Saved experiments
📊

Lead Evaluation

Every candidate gets a report card — validity, affinity, selectivity, drug-likeness, ADME, toxicity, synthesis, developability — with hard gates and Pareto tradeoffs, explained in plain English.

Multi-objective · Hard gates · Pareto
⚛️

Quantum Lab

Explore active spaces, qubits, Hamiltonians, and VQE convergence — and learn the honest boundary between a teaching simulation and real quantum chemistry.

Educational simulation · Clearly labeled
How It Works

From First Principles
to a Defensible Candidate

Nothing to install. Open the site and start learning.

01

Learn the Science

Begin with the Science Journey — atoms, proteins, disease targets, SMILES, and binding. No chemistry background required, and no jargon left unexplained.

02

Design a Molecule

Pick a real disease target, inspect its 3D structure, and generate candidate molecules with a fine-tuned Gemma model. Every SMILES string is validated with RDKit before it scores.

03

Judge the Tradeoffs

Strong binding is not enough. Read the full report card — selectivity, ADME, toxicity, synthesis, developability — and learn why a beautiful binder still fails a hard gate.

04

Prove It in the Arena

Take on the Discovery Arena, beat the known drug's score, and save your work as a reproducible experiment in your learner log.

Disease Targets

20 Curated Targets
Across 5 Disease Areas

Every target is a real protein involved in human disease, sourced from peer-reviewed research and the RCSB Protein Data Bank.

🦀
Cancer
EGFR · BCR-ABL · BRAF V600E · CDK4/6 · HER2
🦠
Viruses
HIV Protease · Influenza NA · SARS-CoV-2 Mpro · HCV NS5B
🧫
Bacteria
MRSA PBP2a · M. tuberculosis InhA · E. coli Gyrase
🧠
Neurological
AChE (Alzheimer's) · MAO-B (Parkinson's) · NMDA Receptor
❤️
Metabolic
DPP-4 (Diabetes) · HMG-CoA (Cholesterol) · ACE (Hypertension)
Comparison

Why Deep2Lead?

Textbooks Web simulators Deep2Lead
Real protein structures⚠️ Static images✅ Live RCSB + AlphaFold
AI molecule generation✅ Fine-tuned Gemma LLM
Drug-likeness validation✅ RDKit, QED, Lipinski
Multi-objective tradeoffs⚠️ Described❌ One score✅ Report card + hard gates
Gamified learning⚠️ Limited✅ XP, battles, badges
Plain-English explanations⚠️ Dense⚠️ Minimal✅ Student-friendly
Honest about its limits❌ Overclaims✅ Simulations labeled as such
Cost💰 Expensive💰 Subscription✅ Free core curriculum
Research Lineage

From research project
to learning platform

Deep2Lead builds on the original 2021 paper — an independent research project by Tarun Kumar Chawdhury (arXiv:2108.05183) — and expands its accessibility mission for the generative AI era.

Deep2Lead (2021)

The original paper introduced accessible drug discovery using a Variational Autoencoder for molecular generation and DeepPurpose DTI for binding affinity — the first tool of its kind requiring no programming.

arXiv:2108.05183 · Independent research
Read the paper (PDF) →

Deep2Lead (2026)

Tanisha Chawdhury's gamification insight transformed the pipeline: replace VAEs with a fine-tuned Gemma 4 LLM, swap DeepPurpose for MAMMAL DTI, and wrap it all in a 3D game any student can play.

Fine-tuned on 225,060 real drug-target pairs
Read the 2026 paper →

What Changed

VAE → Gemma 4 LLM (3× validity, 5× QED). DeepPurpose → MAMMAL DTI (pKd scoring). Static tool → gamified platform with XP, badges, and 7 disease bosses. Expert-only → any student.

100% SMILES validity · QED 0.591 avg
Enter the Arena →
Independent research. Deep2Lead was conceived and built by Tarun Kumar Chawdhury and Tanisha Chawdhury in a personal capacity through DLYog Lab. This work is not affiliated with, endorsed by, or conducted on behalf of any employer or academic institution. Read full disclosure →
Privacy

Your Work Is Yours

Deep2Lead is a hosted platform, so here is plainly what that means for your data.

📥

What We Store

Your account email, your saved experiments and molecule designs, your Arena sessions, and your learning progress — so your work is still there when you come back.

🚫

No Ad Tracking

No advertising trackers and no third-party analytics profiling. We do not sell your data, and we do not share your designs with anyone.

🔑

Passwordless Sign-in

Sign in with a magic link sent to your email — there is no password for us to store or for anyone to leak. Delete your account and your work goes with it.

Important Notice

Educational Use Only

⚠️

Deep2Lead exists exclusively to teach. It is built to make drug discovery and the role of AI in medicine understandable and exciting for students — not to produce medicines.

🚫
Not Medical Advice

Nothing here constitutes medical advice, diagnosis, or treatment. Never use any output from Deep2Lead to make health decisions. Always consult a qualified medical professional.

🧪
Not a Replacement for Lab Research

Molecules generated here are AI-assisted simulations for learning only. They have not been synthesised, tested, validated, or approved by any regulatory authority. Real drug discovery takes years of rigorous laboratory and clinical work.

🤖
AI Limitations

The language model may produce inaccurate, incomplete, or nonsensical structures. Drug-likeness, ADME, toxicity, patient, and quantum values are computational estimates or labeled simulations — they do not predict real-world activity, safety, or efficacy.

🔬
Protein & Structure Data

Protein structures come from RCSB PDB and AlphaFold DB for educational purposes and are subject to their own terms of use. Deep2Lead claims no ownership of any biological data.

⚖️
No Warranty

Deep2Lead is provided "as is" without warranty of any kind, express or implied, including warranties of merchantability, fitness for a particular purpose, or non-infringement. Use at your own risk.

👶
Intended Audience

Designed for students aged 13 and above with appropriate supervision. It is a tool for sparking interest in STEM — not a professional research platform.

Get Started

Start Discovering Today

Free core curriculum, nothing to install. Begin with the science and work your way to a candidate you can defend.

Get In Touch

Questions?

Whether you're a student, educator, or researcher — we'd love to hear from you. Report bugs, suggest new targets, or ask about using Deep2Lead in your classroom.