Research map / Graph learning / Graph self-supervised learning
Learning the Neighborhood: Contrast-Free Multimodal Self-Supervised Molecular Graph Pretraining
Boshra Ariguib, Mathias Niepert, Andrei Manolache
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 DeepSets subgraph aggregation to the 2D-only probe | C-FREE 2D · MoleculeNet linear probe | ROC-AUC avg: 66.63 -> 67.27 | Improved |
| add 3D conformers to the 2D backbone (multimodal, whole-molecule probe) | C-FREE · MoleculeNet linear probe | ROC-AUC avg: 66.63 -> 71.07 | Improved |
| add DeepSets subgraph aggregation to the multimodal probe | C-FREE MM · MoleculeNet linear probe | ROC-AUC avg: 71.07 -> 70.92 | No clear effect |
| use PaiNN encoder with three conformers instead of SchNet single conformer | C-FREE · MoleculeNet full fine-tuning | ROC-AUC avg: 76.63 -> 79.81 | Improved |
| add pretraining to PaiNN on QM9 | PaiNN · QM9 Alpha | MAE: 0.149 -> 0.107 | Improved |
| add pretraining to PaiNN on QM9 Mu | PaiNN · QM9 Mu | MAE: 0.042 -> 0.057 | Hurt |
| add chemically informed Murcko targets instead of EgoNets | C-FREE 3D · QM9 | MAE: 1.155 -> 1.135 | Mixed |
| add self-supervised pretraining at 1% labels | Drugs-75K model · Drugs-75K IP | MAE: 0.638 -> 0.608 | Improved |
| add self-supervised pretraining at 100% labels | Drugs-75K model · Drugs-75K IP | MAE: 0.419 -> 0.419 | No clear effect |
| add 2D modality to 3D-only backbone (ablation) | C-FREE fine-tuned · Kraken L | MAE: 0.329 -> 0.306 | Improved |
| add a transformer predictor (vs none) (ablation) | C-FREE 2D · Kraken B5 linear probe | MAE: 1.065 -> 0.588 | Improved |
| add a transformer predictor (vs MLP) (ablation) | C-FREE 2D · Kraken B5 linear probe | MAE: 0.817 -> 0.588 | Improved |
Closest papers by meaning
- 3D Denoisers Are Good 2D Teachers: Molecular Pretraining via Denoising and Cross-Modal Distillation (AAAI 2025 · similarity 0.93)
- Learning Over Molecular Conformer Ensembles: Datasets and Benchmarks (ICLR 2024 · similarity 0.90)
- Empowering Active Learning for 3D Molecular Graphs with Geometric Graph Isomorphism (NeurIPS 2024 · similarity 0.90)
- Beyond Atoms: Enhancing Molecular Pretrained Representations with 3D Space Modeling (ICML 2025 · similarity 0.89)
- Enhancing Molecular Representation Learning Through the Combination of 3D and 2D Graph Machine Learning (Student Abstract) (AAAI 2025 · similarity 0.89)
- Lift Your Molecules: Molecular Graph Generation in Latent Euclidean Space (ICLR 2025 · similarity 0.89)
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.
