GNN Drug Discovery Pipeline

Overview:

This is my project exploring the use of Graph Neural Networks (GNNs) for predicting key drug discovery properties like lipophilicity, solubility, and blood-brain barrier permeability.
I ran into a lot of issues getting RDKit to work locally (NumPy and version conflicts), so I used Claude to help generate the molecular graph representations. Designed to run in Google Colab.

Approach:

Instead of building separate models for each property, this project uses multi-task learning where one model learns shared chemical knowledge that helps with all predictions. The architecture includes:

-Graph Attention Networks (GAT) for molecular understanding -Shared representation learning across tasks -Task-specific prediction heads for each property -Real pharmaceutical datasets from DeepChem and ChEMBL

• notebooks/ — Colab notebook with full pipeline
• data/ — input data (add your own)
• results/ — output plots and metrics
• models/ — saved models

Requirements:

Python 3.9

Install dependencies with:

pip install -r requirements.txt

Clone this repo:

```bash
git clone https://github.com/<your-username>/GNN-Drug-Discovery.git