Home▸Machine Learning & Neural Networks▸Graph Neural Networks: Learning on Relational Data

🕸️ Graph Neural Networks: Learning on Relational Data

Explore graph neural networks, the message-passing architectures that let neural networks learn directly from nodes, edges and neighborhoods in irregular, relational data like molecules and social networks.

Machine Learning & Neural Networks3DModerate60 FPS

The 3D simulation visualizes a graph of connected nodes exchanging messages layer by layer, showing how each node's embedding is progressively reshaped by aggregating information from its expanding neighborhood as message-passing rounds accumulate.

🔬 What It Demonstrates

🎮 How to Use

Pick a sample graph, choose an aggregation function (sum, mean, or max), and step through message-passing layers one at a time to watch each node's color and position shift as it absorbs information from farther and farther across the graph.

💡 Did You Know?

DeepMind and academic labs have used GNNs to predict molecular properties and even protein interactions at a fraction of the computational cost of traditional physics-based simulations, helping narrow millions of candidate compounds down to a shortlist worth testing in a real lab.

⚙ Under the hood

Watch information spread hop by hop across a graph as each round of message passing lets every node aggregate its neighbors' features.

graph neural networkgnnmessage passingmachine-learninggraphsmolecular property prediction

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install