Research map / Graph learning / Graph self-supervised learning
Conditional Graph Information Bottleneck for Molecular Relational Learning
Namkyeong Lee, Dongmin Hyun, Gyoung S. Na, Sungwon Kim, Junseok Lee, Chanyoung Park
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 conditional compression conditioned on the paired graph instead of single-graph compression (ablation) | CGIB · Chromophore Absorption and Emission | RMSE | Improved |
| Use conditional compression rather than joint mutual-information compression of both graphs (ablation) | CGIB · Chromophore Absorption and Emission | model performance | Improved |
| Add the paired-graph information-preservation term (ablation) | CGIB · Chromophore Absorption and Emission | model performance | Improved |
| Use single-graph subgraph compression instead of no IB (ablation) | relational learning model · Chromophore Absorption and Emission | model performance | Improved |
| Set beta to 1 for aggressive compression (ablation) | CGIB · Emission | model performance | Hurt |
| Set beta to 0 and remove information compression (ablation) | CGIB · Emission | model performance | Hurt |
| Add minimization of mutual information between the target and the paired graph (ablation) | CGIB · Molecular interaction prediction | model performance | Hurt |
| Use asymmetric loss with the task-critical molecule as G1 rather than symmetric loss (ablation) | CGIB · Chromophore Absorption | RMSE: 21.11 -> 17.84 | Improved |
| Use asymmetric loss rather than symmetric loss (ablation) | CGIB · Chromophore Emission | RMSE: 25.49 -> 24.44 | Improved |
| Use asymmetric loss rather than symmetric loss (ablation) | CGIB · Chromophore Lifetime | RMSE: 0.798 -> 0.796 | Improved |
| Use contrastive learning instead of the variational IB-based paired-graph term (ablation) | CGIB · FreeSolv | RMSE: 0.831 -> 0.852 | Hurt |
| Use contrastive learning instead of the variational IB-based paired-graph term (ablation) | CGIB · ZhangDDI transductive | AUROC: 94.74 -> 93.78 | Hurt |
Closest papers by meaning
- Pre-Training Graph Neural Networks on Molecules by Using Subgraph-Conditioned Graph Information Bottleneck (AAAI 2025 · similarity 0.91)
- Neural Atoms: Propagating Long-range Interaction in Molecular Graphs through Efficient Communication Channel (ICLR 2024 · similarity 0.88)
- Reaction Graph: Towards Reaction-Level Modeling for Chemical Reactions with 3D Structures (ICML 2025 · similarity 0.86)
- Empowering Active Learning for 3D Molecular Graphs with Geometric Graph Isomorphism (NeurIPS 2024 · similarity 0.86)
- Enhancing Molecular Representation Learning Through the Combination of 3D and 2D Graph Machine Learning (Student Abstract) (AAAI 2025 · similarity 0.86)
- MORE: Molecule Pretraining with Multi-Level Pretext Task (AAAI 2025 · 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.
