Research map / Graph learning / Graph self-supervised learning
Molecular Representation Learning via Heterogeneous Motif Graph Neural Networks
Zhaoning Yu, Hongyang Gao
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add heterogeneous motif-level representations to atom-level GIN (ablation) | HM-GNN · PTC, MUTAG, PROTEINS | accuracy: 64.6 $\pm$ 7.0 -> 78.8 $\pm$ 6.5; 89.4 $\pm$ 5.6 -> 96.3 $\pm$ 2.6; 76.2 $\pm$ 2.8 -> 79.9 $\pm$ 3.1 | Improved |
| Add motif-motif interactions to the heterogeneous motif graph (ablation) | HM-GNN · PTC, MUTAG, PROTEIN | accuracy: 76.4$\pm$4.4 -> 78.8$\pm$6.5; 94.6$\pm$4.3 -> 96.3$\pm$2.6; 78.3$\pm$3.5 -> 79.9$\pm$3.1 | Improved |
| Combine PTC with PTC_MM and shared motif vocabulary (ablation) | HM-GNN multi-task learning · PTC, 90% training | accuracy: 71.8 $\pm$ 4.1 -> 76.5 $\pm$ 3.3 | Improved |
| Combine PTC with PTC_MM and shared motif vocabulary (ablation) | HM-GNN multi-task learning · PTC, 50% training | accuracy: 65.1 $\pm$ 0.8 -> 69.2 $\pm$ 0.8 | Improved |
| Combine PTC with PTC_MM and shared motif vocabulary (ablation) | HM-GNN multi-task learning · PTC, 10% training | accuracy: 59.9 $\pm$ 1.9 -> 66.7 $\pm$ 1.9 | Improved |
| Combine PTC with PTC_FR and shared motif vocabulary (ablation) | HM-GNN multi-task learning · PTC, 90% training | accuracy: 71.8 $\pm$ 4.1 -> 84.3 $\pm$ 3.8 | Improved |
| Combine PTC with PTC_FR and shared motif vocabulary (ablation) | HM-GNN multi-task learning · PTC, 50% training | accuracy: 65.1 $\pm$ 0.8 -> 77.3 $\pm$ 0.8 | Improved |
| Combine PTC with PTC_FR and shared motif vocabulary (ablation) | HM-GNN multi-task learning · PTC, 10% training | accuracy: 59.9 $\pm$ 1.9 -> 74.0 $\pm$ 1.7 | Improved |
| Use HM-GNN instead of the strongest listed baseline | TUDataset graph classification · PTC, MUTAG, NCI1, PROTEINS, MUTAGENICITY | accuracy | Improved |
| Use HM-GNN rather than PNA | ogbg-molhiv · ogbg-molhiv | ROC-AUC: 79.05 $\pm$ 1.32 -> 79.03 $\pm$ 0.92 | No clear effect |
| Add PNA to HM-GNN | HM-GNN · ogbg-molhiv | ROC-AUC: 79.03 $\pm$ 0.92 -> 80.20 $\pm$ 1.18 | Improved |
| Increase the motif vocabulary keeping ratio beyond 80% (ablation) | HM-GNN · ogbg-molhiv | ROC-AUC | Hurt |
Closest papers by meaning
- Towards Multiscale Graph-based Protein Learning with Geometric Secondary Structural Motifs (NeurIPS 2025 · similarity 0.89)
- MAGE: Model-Level Graph Neural Networks Explanations via Motif-based Graph Generation (ICLR 2025 · similarity 0.89)
- Robust Heterogeneous Graph Classification for Molecular Property Prediction with Information Bottleneck (AAAI 2025 · similarity 0.89)
- Pre-Training Graph Neural Networks on Molecules by Using Subgraph-Conditioned Graph Information Bottleneck (AAAI 2025 · similarity 0.88)
- On the Scalability of GNNs for Molecular Graphs (NeurIPS 2024 · similarity 0.87)
- A Simple and Scalable Representation for Graph Generation (ICLR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
