Research map / Graph learning / Graph self-supervised learning
Simple GNN Regularisation for 3D Molecular Property Prediction and Beyond
Jonathan Godwin, Michael Schaarschmidt, Alexander L Gaunt, Alvaro Sanchez-Gonzalez, Yulia Rubanova, Petar Veličković, James Kirkpatrick, Peter Battaglia
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 |
|---|---|---|---|
| Add Noisy Nodes to GNS | GNS · OC20 IS2RE test OOD Both | eV MAE: 0.638 -> 0.465 | Improved |
| Add Noisy Nodes to GNS | GNS · OC20 IS2RS test OOD Both | ADwT: 46.9% -> 52.7% | Improved |
| Add Noisy Nodes to GNS | GNS · QM9 test | std. MAE: 1.17 -> 0.90 | Improved |
| Add noise but omit the node-level correction target (ablation) | GNS · QM9 test | std. MAE: 1.17 -> 1.16 | No clear effect |
| Use the invariance variant (ablation) | GNS-10 + Noisy Nodes · QM9 test | std. MAE: 0.88 -> 0.92 | Hurt |
| Increase depth from 10 to 30 layers (ablation) | GNS-10 + Noisy Nodes · QM9 test | std. MAE: 0.89 -> 0.88 | Improved |
| Interpolate between initial and final positions rather than sample trajectory states (ablation) | GNS + Noisy Nodes · OC20 IS2RE ID validation | eV MAE: 0.47 -> 0.45 | Improved |
| Add Noisy Nodes to 50-layer MPNN | MPNN + Virtual Node · OGBG-PCQM4M | MAE: 0.1236 -> 0.1218 | Improved |
| Add Noisy Nodes to 16-layer MPNN | MPNN + Virtual Node · OGBG-MOLPCBA | Mean AP: 27.6% ± 0.004 -> 28.1% ± 0.002 | Improved |
| Add Noisy Nodes to 4-layer GCN with residual connections | 4-layer GCN · OGBN-Arxiv | accuracy: 72.39% ± 0.002 -> 72.52% ± 0.003 | Improved |
| Set category-flip probability to 0.1 instead of 0.05 (ablation) | 8-layer MPNN + Virtual Node + Noisy Nodes · OGBG-MOLPCBA | Mean AP: 28.1% +- 0.001 -> 28.0% +- 0.003 | Hurt |
| Use DropEdge instead of Noisy Nodes (ablation) | MPNN · OGBG-MOLPCBA | Mean AP: 27.4% ± 0.002 -> 27.5% ± 0.001 | No clear effect |
Closest papers by meaning
- Empowering Active Learning for 3D Molecular Graphs with Geometric Graph Isomorphism (NeurIPS 2024 · similarity 0.86)
- On the Scalability of GNNs for Molecular Graphs (NeurIPS 2024 · similarity 0.86)
- On the Generalization of Equivariant Graph Neural Networks (ICML 2024 · similarity 0.86)
- Robust Heterogeneous Graph Classification for Molecular Property Prediction with Information Bottleneck (AAAI 2025 · similarity 0.86)
- 3D Denoisers Are Good 2D Teachers: Molecular Pretraining via Denoising and Cross-Modal Distillation (AAAI 2025 · similarity 0.85)
- Neural Atoms: Propagating Long-range Interaction in Molecular Graphs through Efficient Communication Channel (ICLR 2024 · similarity 0.85)
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.
