Estrogen Receptor Blocker

About 70%-80% of breast cancers have estrogen receptors. Our platform is designed to generate, evaluate, and optimize novel drug candidates targeting the human estrogen receptor while minimizing toxicity.

Methods

1. Generate possible drugs - most drugs that inhibit the breast cancer target share a core chemical backbone called a triphenylethylene core or a phenol group. A partial molecule input along with some configurations are passed to ChemGPT-4.7M to generate possible molecules.
2. Optimize existing drugs - a complete molecule can be slightly transformed using rdChemReactions based on our list of possible transformations to attempt to get better measurements.

Every molecule generated/optimized are passed into RDKit.Chem to analyze the molecular weight, LogP, HBD/HBA, and synthetic accessibility. Only molecules that pass the filters for these metrics are shown as results.

The toxicity and ER-alpha activity values are predicted using Random Forest Classifier from these datasets:

1. ClinTox includes drugs approved by the FDA and those that have failed clinical trials for toxicity reasons. We use this dataset to predict the toxicity value.
2. CHEMBL206 includes bioacticity records of molecules that interact with the Estrogen receptor alpha. We use this dataset to predict the ERalpha activity value.

The training information is available in metrics.json or using the "View model training info" button on webpage.

Tech Stack

The client web UI (HTML/CSS/JS) sends HTTPS REST requests to a Python Rust/FastAPI server. The backend is fully written in Python, using rdkit, scikit-learn, and other libraries.

Running the Project

Install the Python dependencies"

python3 -m venv .venv
source .venv/bin/activate
python3 -m pip install -r requirements.txt

If you want to train the model:

rm ./artifacts/*
python3 train_models.py

If you want to use our pre-trained artifacts, they are already included in artifacts, no action needed.

Start the application:

python3 -m uvicorn api:app --host 127.0.0.1 --port 8000

Open http://localhost:8000. API doc is at http://localhost:8000/docs.

Environment Variables

In the .env file (not included in this repository), these values are required:

• HF_TOKEN, a Hugging Face access token