Research map / Graph learning / Graph self-supervised learning
Rethinking Tokenizer and Decoder in Masked Graph Modeling for Molecules
Zhiyuan Liu, Yaorui Shi, An Zhang, Enzhi Zhang, Kenji Kawaguchi, Xiang Wang, Tat-Seng Chua
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| Use SimSGT pretraining rather than the strongest listed baseline | GINE · MoleculeNet (8 datasets) | ROC-AUC (%) average: 74.0 -> 74.4 | Improved |
| Use SimSGT pretraining rather than the strongest listed baseline | GraphTrans · MoleculeNet (8 datasets) | ROC-AUC (%) average: 73.4 -> 75.8 | Improved |
| Use SimSGT rather than the strongest listed baseline | GTS encoder · QM7 | MAE: 78.4 -> 75.4 | Improved |
| Use SimSGT rather than the strongest listed baseline | GTS encoder · QM8 | MAE: 0.0200 -> 0.0183 | Improved |
| Use SimSGT rather than the strongest listed baseline | GTS encoder · QM9 | MAE: 5.75 -> 5.53 | Improved |
| Use SimSGT rather than GraphMAE for pretraining | GTS encoder on ZINC15 · ZINC15, 100 pretraining epochs | pretraining time: 527 min -> 645 min | Hurt |
| Use SimSGT rather than Mole-BERT for pretraining | GTS encoder on ZINC15 · ZINC15, 100 pretraining epochs | pretraining time: 2199 min -> 645 min | Improved |
| replace SimSGT pooling with mean pooling (ablation) | SimSGT · MoleculeNet (8 datasets), GTS encoder | ROC-AUC (%) average: 75.8 -> 75.2 | Hurt |
| replace SimSGT pooling with max pooling (ablation) | SimSGT · MoleculeNet (8 datasets), GTS encoder | ROC-AUC (%) average: 75.8 -> 74.5 | Hurt |
| replace SimSGT pooling with sum pooling (ablation) | SimSGT · MoleculeNet (8 datasets), GTS encoder | ROC-AUC (%) average: 75.8 -> 74.8 | Hurt |
Closest papers by meaning
- 3D-GSRD: 3D Molecular Graph Auto-Encoder with Selective Re-mask Decoding (NeurIPS 2025 · similarity 0.89)
- Bridging Molecular Graphs and Large Language Models (AAAI 2025 · similarity 0.88)
- Graph Generative Pre-trained Transformer (ICML 2025 · similarity 0.88)
- Pre-Training Graph Neural Networks on Molecules by Using Subgraph-Conditioned Graph Information Bottleneck (AAAI 2025 · similarity 0.88)
- Self-supervised Masked Graph Autoencoder via Structure-aware Curriculum (ICML 2025 · similarity 0.87)
- GraphGPT: Generative Pre-trained Graph Eulerian Transformer (ICML 2025 · similarity 0.87)
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.
