This repository provides the official PyTorch implementation of E(n) Equivariant Graph Neural Networks (EGNNs), a novel model designed for learning graph neural networks that are equivariant to rotations, translations, reflections, and permutations. It targets researchers and practitioners in fields like dynamical systems modeling, representation learning, and molecular property prediction, offering competitive performance without requiring computationally expensive higher-order representations.

How It Works

EGNNs achieve equivariance by leveraging a unique architecture that processes node features and coordinates separately. This approach allows for efficient computation and scalability to higher-dimensional spaces, unlike prior methods limited to 3D. The model's design enables it to maintain geometric consistency across transformations, leading to improved accuracy and generalization on tasks involving physical systems and molecular structures.

Quick Start & Requirements

• Install: Copy models/egnn_clean/egnn_clean.py into your working directory.
• Prerequisites: PyTorch 1.7.1.
• Example: The README provides a runnable PyTorch snippet demonstrating EGNN initialization and forward pass.
• Experiments: Scripts are available for N-body systems, graph autoencoders, and QM9 molecular property prediction.

Highlighted Details

• Achieves competitive or better performance than existing methods without higher-order representations.
• Easily scales to higher-dimensional spaces beyond 3D.
• Demonstrated effectiveness on dynamical systems modeling, graph autoencoders, and molecular property prediction.
• Includes example code and scripts for various benchmark experiments.

Maintenance & Community

The project acknowledges financial support from Robert Bosch GmbH. Further community interaction channels are not explicitly mentioned in the README.

Licensing & Compatibility

The repository does not explicitly state a license. This may pose a restriction for commercial use or integration into closed-source projects.

Limitations & Caveats

The README specifies PyTorch 1.7.1, which is an older version and may require careful dependency management. The absence of an explicit license is a significant caveat for adoption.