Research map / Graph learning / Graph self-supervised learning
MolGrow: A Graph Normalizing Flow for Hierarchical Molecular Generation
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 |
|---|---|---|---|
| Use fragment-oriented atom ordering rather than BFS (ablation) | MolGrow · MOSES | FCD/Test: 9.962 \pm 0.795 -> 6.284 \pm 0.986 | Helped |
| Use fragment-oriented atom ordering rather than BFS (ablation) | MolGrow · MOSES | Frag/Test: 0.932 \pm 0.01 -> 0.929 \pm 0.025 | No clear effect |
| Use fragment-oriented ordering rather than BFS ordering within fragments (ablation) | MolGrow · MOSES | FCD/Test: 16.15 \pm 1.026 -> 6.284 \pm 0.986 | Helped |
| Use fragment-oriented atom ordering rather than random permutation (ablation) | MolGrow · MOSES | FCD/Test: 40.17 \pm 4.709 -> 6.284 \pm 0.986 | Helped |
| Use CAGE instead of GAT (ablation) | MolGrow · MOSES | FCD/Test: 6.523 \pm 0.302 -> 6.284 \pm 0.986 | Helped |
| Add positional embeddings (ablation) | MolGrow · MOSES | FCD/Test: 6.771 \pm 0.555 -> 6.284 \pm 0.986 | Helped |
| Mutate latent codes across levels rather than only the top level (ablation) | MolGrow (genetic optimization) · Global optimization, ZINC250k | penalized logP (1st): 11.66 \pm 0.31 -> 14.01 \pm 0.364 | Helped |
| Mutate latent codes across levels rather than only the bottom level (ablation) | MolGrow (genetic optimization) · Global optimization, ZINC250k | penalized logP (1st): 10.29 \pm 3.32 -> 14.01 \pm 0.364 | Helped |
| Use MolGrow rather than MoFlow for constrained optimization | Molecular optimization · ZINC250k, δ=0.4 | penalized logP improvement: 4.71 \pm 4.55 -> 8.337 \pm 6.85 | Helped |
| Use MolGrow rather than MoFlow for constrained optimization | Molecular optimization · ZINC250k, δ=0.6 | penalized logP improvement: 2.10 \pm 2.86 -> 4.063 \pm 5.609 | Mixed |
| Use MolGrow rather than MoFlow for distribution learning | MOSES graph generation · MOSES | FCD/Test: 28.05 -> 6.284 \pm 0.986 | Helped |
Closest papers by meaning
- Junction Tree Variational Autoencoder for Molecular Graph Generation (ICML 2018 · similarity 0.90)
- Graph Diffusion that can Insert and Delete (NeurIPS 2025 · similarity 0.88)
- Navigating Chemical Space with Latent Flows (NeurIPS 2024 · similarity 0.88)
- Pretraining Generative Flow Networks with Inexpensive Rewards for Molecular Graph Generation (ICML 2025 · similarity 0.88)
- DMol: A Highly Efficient and Chemical Motif-Preserving Molecule Generation Platform (NeurIPS 2025 · similarity 0.88)
- MolHIT: Advancing Molecular-Graph Generation with Hierarchical Discrete Diffusion Models (NeurIPS 2026 · similarity 0.88)
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.