Research map / Probabilistic and scientific modeling / Molecules and proteins
LeakGFN: Robust Molecular Generation in Generative Flow Networks via Flow Decomposition
Hwanhee Kim, Seungyeon Choi, Sanghyun Park
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 dual-head flow decomposition (valid head) to flow matching (ablation) | GFN-FM · JNK3 | HM: 0.40 -> 0.65 | Improved |
| add flow-guided exploration instead of uniform epsilon-exploration (ablation) | LeakGFN · JNK3 | HM: 0.62 -> 0.65 | Improved |
| add flow decomposition to GFN-FM | GFN-FM · JNK3 | HM std over seeds: 0.218 -> 0.068 | Improved |
| add flow decomposition to GFN-FM | GFN-FM · GSK3β | HM: 0.716 -> 0.760 | Improved |
| add flow decomposition to GFN-FM | GFN-FM · QED | HM: 0.926 -> 0.926 | No clear effect |
| use LeakGFN instead of balance-based objectives (DB) at Lmax = 8 | GFlowNet · DRD2 | HM: 0.934 -> 0.797 | Hurt |
| increase Lmax from 8 to 12 for balance-based GFN-DB | GFN-DB · DRD2 | HM: 0.934 -> 0.096 | Hurt |
| increase Lmax from 8 to 12 for LeakGFN | LeakGFN · DRD2 | HM: 0.797 -> 0.435 | Hurt |
| set R(x)=0 at forced terminals under flow matching (ablation) | GFN-FM · JNK3 | HM: 0.403 -> 0.122 | Hurt |
| set R(x)=0 at forced terminals under trajectory balance (ablation) | GFN-TB · JNK3, GSK3β, DRD2 | HM: 0.624 -> 0.491 (JNK3); 0.675 -> 0.714 (GSK3β) | Mixed |
| plug LeakGFN dual head into TacoGFN | TacoGFN · CrossDocked | Vina Dock mean: -8.237 -> -8.589 | Improved |
| plug LeakGFN dual head into TacoGFN | TacoGFN · CrossDocked | Diversity mean: 0.530 -> 0.514 | Hurt |
Closest papers by meaning
- Pretraining Generative Flow Networks with Inexpensive Rewards for Molecular Graph Generation (ICML 2025 · similarity 0.91)
- Generative Flows on Synthetic Pathway for Drug Design (ICLR 2025 · similarity 0.90)
- Navigating Chemical Space with Latent Flows (NeurIPS 2024 · similarity 0.90)
- SynFlowNet: Design of Diverse and Novel Molecules with Synthesis Constraints (ICLR 2025 · similarity 0.89)
- Compositional Flows for 3D Molecule and Synthesis Pathway Co-design (ICML 2025 · similarity 0.89)
- Flow Density Control: Generative Optimization Beyond Entropy-Regularized Fine-Tuning (NeurIPS 2025 · 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.
