GNN Libraries logopeterindia.net
Knowledge Hub

Open-source Libraries for Deep Learning on Graphs (GNN)

A directory of open-source libraries for deep learning on graphs, organised by framework — PyTorch, TensorFlow and JAX, and Julia — for tasks from drug discovery to knowledge graph completion.

12 topics•4 categories•Updated October 2026

PyTorch Libraries

4 entries

Libraries built on PyTorch.

01PyTorch Geometric (PyG)A library built upon PyTorch to easily write and train Graph Neural Networks — implementing 50+ GNN operators (GCN, GAT, GraphSAGE, GIN, GATv2, MPNN) with a unified message-passing API, mini-batch support, and data loaders for common graph datasets including OGB. torch-geometric 2.8.1 was released on 12 October 2026, according to PyPI. 02Deep Graph Library (DGL)Easy deep learning on graphs — a framework-agnostic library (PyTorch, TensorFlow, MXNet) implementing dozens of GNN models including GCN, GraphSAGE, GAT, and GIN. Features efficient sparse message-passing kernels and large-scale graph support with DGL-KE and DGL-LifeSci. The latest PyPI release (dgl 2.2.1) is from May 2024 and the latest GitHub release listed (v2.4.0) is from June 2024, with the repository last updated in July 2025. 03TorchDrugPyTorch-based platform for drug discovery with graph neural networks. The latest PyPI release (0.2.1) is from July 2023, and the repository was last updated in August 2024. 04ptgnn (repository archived)Microsoft Research's PyTorch graph neural network library — providing composable GNN components, training infrastructure, and implementations of multiple GNN architectures used internally for tasks including code understanding and program analysis.

TensorFlow & JAX Libraries

3 entries

Libraries for TensorFlow and JAX users.

05SpektralA Python library for graph deep learning based on the Keras API and TensorFlow 2 — supporting graph classification, node classification, link prediction, and graph generation. Integrates natively with Keras layers and the tf.data pipeline. The latest PyPI release (1.3.1) and the repository's last update are both from January 2024. 06Graph NetsGoogle DeepMind's library for building graph networks in TensorFlow and Sonnet — implementing the general Graph Networks framework (Battaglia et al.) supporting arbitrary edge, node, and global computation with flexible message-passing schemes. The latest PyPI release (1.1.0) is from January 2020, and the repository's last commit was in December 2022. 07Jraph (repository archived)A lightweight library for working with graph neural networks in JAX — defining a GraphsTuple data structure and providing utilities for batching graphs, padding, and implementing message-passing GNN layers with JAX's functional, composable style.

Julia Libraries

1 entries

Graph deep learning in Julia.

08GeometricFluxThe geometric deep learning library for Flux (Julia) — providing classic GNN layers including graph convolutional networks, graph attention networks, and message passing neural networks. Compatible with the JuliaGraphs ecosystem and CUDA GPU acceleration. The latest GitHub release listed (v0.14.0) is from May 2023, and the repository was last updated in October 2023.

Related Pages

4 entries

Related directories on this site.

09GNN FrameworksGNN frameworks and graph tools. 10Graph Neural NetworksGCN, GAT, GraphSAGE and GIN explained. 11Deep Learning FrameworksDeep learning frameworks. 12Knowledge Graphs & Graph TechnologiesGraph databases and knowledge graphs.

Frequently Asked Questions

What is a GNN library?

A library that implements graph neural network layers, message passing and graph data handling on top of a deep learning framework.

Which library is most widely used?

PyTorch Geometric is the most actively released: version 2.8.1 came out on 12 October 2026, according to PyPI. Deep Graph Library is widely used, but its latest PyPI and GitHub releases are from 2024.

Which library fits TensorFlow or JAX?

Spektral targets TensorFlow and Keras; Graph Nets and Jraph target TensorFlow and JAX, but the Jraph repository is archived.

Are all of these maintained?

No. The Jraph and ptgnn repositories are archived. Several others have not had a release in years: Spektral (January 2024), TorchDrug (July 2023), GeometricFlux (May 2023) and Graph Nets (PyPI, January 2020). Check recent activity before choosing a library.

Where are general graph tools?

See the GNN Frameworks page for NetworkX, igraph and other graph analysis libraries.