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.
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.
Graph select, aggregation function select, layer step control, play/pause, rebuild
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.
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.
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.
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.
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.