Research map / Graph learning / Graph self-supervised learning
Pre-training Molecular Graph Representation with 3D Geometry
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Pre-train the 2D encoder with paired 3D geometry using contrastive and generative SSL | GIN molecular encoder · Molecular property prediction (8 tasks, Avg) | ROC-AUC: 67.21 -> 71.69 | Helped |
| Add 2D generative SSL (AttrMask) to GraphMVP | GraphMVP · Molecular property prediction (8 tasks, Avg) | ROC-AUC: 71.69 -> 72.76 | Helped |
| Add 2D contrastive SSL (ContextPred) to GraphMVP | GraphMVP · Molecular property prediction (8 tasks, Avg) | ROC-AUC: 71.69 -> 73.07 | Helped |
| Add 3D geometry pre-training | GIN molecular encoder · 3D Diameter Prediction | accuracy: 38.9±0.8 -> 42.3±1.3 | Helped |
| Add 3D geometry pre-training | GIN molecular encoder · Long-Range Donor-Acceptor Detection | accuracy: 77.9±1.1 -> 81.5±0.4 | Helped |
| Increase masking ratio from M=0 to M=0.15 (ablation) | GraphMVP · Molecular property prediction (8 tasks, Avg) | ROC-AUC: 71.12 -> 71.60 | Helped |
| Increase masking ratio from M=0.15 to M=0.30 (ablation) | GraphMVP · Molecular property prediction (8 tasks, Avg) | ROC-AUC: 71.60 -> 71.79 | Helped |
| Increase conformers per molecule from C=1 to C=5 (ablation) | GraphMVP · Molecular property prediction (8 tasks, Avg) | ROC-AUC: 71.61 -> 71.60 | No clear effect |
| Increase conformers per molecule from C=5 to C=10 (ablation) | GraphMVP · Molecular property prediction (8 tasks, Avg) | ROC-AUC: 71.60 -> 72.20 | Helped |
| Increase conformers per molecule from C=10 to C=20 (ablation) | GraphMVP · Molecular property prediction (8 tasks, Avg) | ROC-AUC: 72.20 -> 72.39 | Helped |
| Add InfoNCE contrastive SSL (ablation) | 3D-pretrained 2D molecular encoder · Molecular property prediction (8 tasks, Avg) | ROC-AUC: 67.21 -> 68.85 | Helped |
| Add EBM-NCE contrastive SSL (ablation) | 3D-pretrained 2D molecular encoder · Molecular property prediction (8 tasks, Avg) | ROC-AUC: 67.21 -> 70.15 | Helped |
Closest papers by meaning
- Enhancing Molecular Representation Learning Through the Combination of 3D and 2D Graph Machine Learning (Student Abstract) (AAAI 2025 · similarity 0.92)
- 3D Denoisers Are Good 2D Teachers: Molecular Pretraining via Denoising and Cross-Modal Distillation (AAAI 2025 · similarity 0.90)
- Empowering Active Learning for 3D Molecular Graphs with Geometric Graph Isomorphism (NeurIPS 2024 · similarity 0.89)
- HAGO-Net: Hierarchical Geometric Message Passing for Molecular Representation Learning (AAAI 2024 · similarity 0.88)
- 3D-GSRD: 3D Molecular Graph Auto-Encoder with Selective Re-mask Decoding (NeurIPS 2025 · similarity 0.88)
- Triplet Interaction Improves Graph Transformers: Accurate Molecular Graph Learning with Triplet Graph Transformers (ICML 2024 · 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.