8 subgraph classification datasets including HPO-NEURO, HPO-METAB, and PPI-BP, designed to evaluate how neural networks capture subgraph topology. The collection includes subgraphs of varying sizes and connectivity patterns embedded within large base graphs representing biological and synthetic networks.
Use Cases
- Train subgraph-level classifiers to predict phenotype categories using the HPO-NEURO and HPO-METAB label sets.
- Analyze the performance of graph neural networks on capturing subgraph-to-graph boundary edges.
- Develop embedding techniques that distinguish between internal subgraph structure and external neighborhood context.
Strengths
- Includes 8 distinct datasets with subgraphs ranging from small clusters to large, sparse components.
- Features ground-truth labels for biological pathways and phenotype ontologies like HPO-NEURO and HPO-METAB.
- Provides adjacency matrices and node features for both the subgraphs and their surrounding base graphs to test boundary-aware models.