It started with a diagnosis.
I'm Sohaib Khan. I was diagnosed with oral squamous cell carcinoma — “SSC”. Hearing those words changes how you see everything. I went looking for answers, and what I found was a field doing extraordinary science behind expensive paywalls, scattered databases, and command-line tools that most people can never touch.
So I built NaturaScreen: an open platform that puts real compound screening, neoantigen targeting, and tumor simulation into one place anyone can run. The goal isn't to claim a cure — it's to help researchers, students, and the curious test more ideas, faster, against cancer, with honest tools and open data.
I can't fight this alone, and neither can any one lab. But thousands of people screening open compounds against real targets, sharing what holds up in the dish — that's a force. This is my contribution to it.
What this is — and what it is not
NaturaScreen produces research hypotheses for the lab. It is not a treatment, a cure, a dose, or medical advice, and nothing it outputs is validated for human use. A compound that looks promising in a model still needs cell-culture work, animal studies, and human clinical trials. If you or someone you love is facing cancer, please work with qualified clinicians. The honesty is the point — it's wired into the code so it can never be turned off.
How it helps researchers
- Open compound library
Hundreds of thousands of natural products from COCONUT (CC0) — searchable, with structures and descriptors.
- Neoantigen targeting
Predict tumor-specific peptide–MHC presentation (MHCflurry) — the targets personalized vaccines aim at — and point compounds at them.
- Molecular docking
Score how strongly a compound binds a cancer target (AutoDock Vina) within a curated pocket.
- Response prediction
An ML model (XGBoost on GDSC) estimates cell-line potency — with honest out-of-distribution flags for natural products.
- Live tumor simulation
Watch an agent-based tumor respond in real time — an illustration of the score, never a prediction.
- Lab-result feedback
Real assay results flow back in and retrain the model, so the ranking sharpens against reality over time.
The potential
Every result is a candidate to investigate, not a conclusion. But the search space of nature is enormous, and most of it has never been screened against most tumors. A free, open, honest tool lowers the cost of asking “what if?” — and the long road ahead points toward screening against a model of one patient's own tumor, with a robotic lab confirming the best picks on living tissue. That last mile — a real human, over real time — is exactly why clinical trials exist and why this tool will never replace them. It aims to make the candidate that reaches a trial far more likely to work.
Run it yourself
It's open source. Bring up the whole stack locally with Docker:
git clone https://github.com/sohaibwcws/naturascreen && cd naturascreen cp .env.example .env make up # postgres, redis, api, worker, web make migrate make ingest-compounds n=300 # live, CC0 natural products open http://localhost:3000
Full setup, the scientific adapters, and the production (HTTPS) deploy are in the project README.
Support the mission — help fight cancer
NaturaScreen is free and open, built in the open so the whole community can improve it. The best way to help is to use it, break it, and make it better:
- ★ Star & share the repository — reach is how open science compounds.
- ⚙ Contribute — better models, curated targets, validated lab results, new datasets.
- 🔬 Bring real data — submit lab results so the model learns what holds up in the dish.
- 📣 Tell a researcher who could put it to work.