Research map / Probabilistic and scientific modeling / Molecules and proteins
Synthesizable Molecular Generation via Soft-constrained GFlowNets with Rich Chemical Priors
Hyeonah Kim, Minsu Kim, Celine Roget, Dionessa Biton, Louis Vaillancourt, Yves V. Brun, Yoshua Bengio, Alex Hernández-García
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 contrastive auxiliary loss on replay buffers (vs reward shaping) (ablation) | RTB post-training of GP-MolFormer · sEH proxy | Positive Top-100 sEH: 1.040 -> 1.043 | Improved |
| replace reward shaping with auxiliary loss (ablation) | RTB post-training · PMO GSK3β | AUC Top-10: 0.502 -> 0.807 | Improved |
| add genetic exploration to S3-GFN (ablation) | S3-GFN · PMO GSK3β | AUC Top-10: 0.807 -> 0.905 | Improved |
| add genetic exploration to reward shaping (ablation) | RTB + RS · PMO GSK3β | AUC Top-10: 0.502 -> 0.862 | Improved |
| remove auxiliary loss (positive-only training) (ablation) | S3-GFN · 2D grid world | probability mass on infeasible region | Hurt |
| use SMILES-based S3-GFN instead of reaction-based GFlowNet | reaction-based GFlowNets · LIT-PCBA ADRB2 | Vina score (kcal/mol): -11.45 -> -12.32 | Improved |
| use S3-GFN instead of reaction-based RxnFlow | RxnFlow · LIT-PCBA ADRB2 | AiZynthFinder success rate (%): 60.25 -> 100.0 | Improved |
| use S3-GFN vs reward shaping under changed constraints (ablation) | RTB + RS · sEH changed constraints | Num. unique: 770.3 -> 929.0 | Improved |
| reward shaping vs S3-GFN positive ratio under changed constraints (ablation) | S3-GFN · sEH changed constraints | Positive ratio: 0.909 -> 0.883 | Hurt |
| use reward shaping with policy-gradient REINVENT | REINVENT · PMO GSK3β | AUC Top-10: 0.830 | Improved |
| pretrained SMILES prior with GFlowNet post-training | GP-MolFormer prior · sEH proxy | Positive ratio: 0.299 -> 0.945 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- SynFlowNet — SynFlowNet: Design of Diverse and Novel Molecules with Synthesis Constraints (ICLR 2025) · better on 8 of 9 shared results
- RGFN — RGFN: Synthesizable Molecular Generation Using GFlowNets (NeurIPS 2024) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SynFlowNet: Design of Diverse and Novel Molecules with Synthesis Constraints (ICLR 2025 · similarity 0.93)
- Generative Flows on Synthetic Pathway for Drug Design (ICLR 2025 · similarity 0.91)
- RGFN: Synthesizable Molecular Generation Using GFlowNets (NeurIPS 2024 · similarity 0.91)
- Pretraining Generative Flow Networks with Inexpensive Rewards for Molecular Graph Generation (ICML 2025 · similarity 0.90)
- Scalable and Cost-Efficient de Novo Template-Based Molecular Generation (NeurIPS 2025 · similarity 0.88)
- Generative Flow Networks for Lead Optimization in Drug Design (Student Abstract) (AAAI 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.
