PhyloGNN: A User-Friendly and Extensible Graph-Learning Toolkit for Fossilized Phylogenies

Fossilized phylogenetic trees contain rich information about evolutionary history, but they remain difficult to use with modern graph-based machine-learning methods. Although the fossilized birth–death (FBD) model has provided an important foundation for integrating fossils, diversification, and sampling into phylogenetic inference, practical workflows for applying graph neural networks (GNNs) to fossilized phylogenies are still often fragmented, highly customized, and difficult to reuse across studies. We introduce PhyloGNN, a Python software toolkit designed to make graph-learning analyses of fossilized phylogenetic data easier to use, easier to reproduce, and easier to extend. PhyloGNN provides an end-to-end workflow for transforming phylogenetic trees into graph representations suitable for deep learning. The toolkit includes reusable components for node-level feature engineering, tree-to-graph conversion, model training, and prediction, allowing users to move from phylogenetic input data to graph-based analyses within a unified framework. It is designed around practical usability: users can run standard workflows with minimal custom code, while the modular structure keeps the framework accessible for more advanced methodological development. PhyloGNN was initially motivated by our previous GAT-LSTM study on simulation-based inference from fossilized phylogenies, but it is designed as a general, user-friendly foundation for graph learning on fossilized phylogenetic data rather than as a single-purpose implementation. Its modular architecture allows users to customize node features, graph construction strategies, pooling and readout schemes, and model backbones, making it straightforward to extend the framework to a wide range of tasks on fossilized phylogenies, including classification, regression, parameter estimation, event detection, and other simulation-based inference problems. By lowering the technical barrier to building, testing, and comparing graph-based workflows, PhyloGNN can serve both as a practical analysis toolkit and as a platform for developing new GNN methods in paleobiology.