Research map / Probabilistic and scientific modeling / Molecules and proteins
FragFM: Hierarchical Framework for Efficient Molecule Generation via Fragment-Level Discrete Flow Matching
Joongwon Lee, Seonghwan Kim, Seokhyun Moon, Hyunwoo Kim, Woo Youn Kim
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 |
|---|---|---|---|
| replace BRICS with RECAP fragmentation (ablation) | FragFM · MOSES | FCD / Scaf / Novelty: FCD 0.58 -> 0.56; Scaf 10.9 -> 13.3; Novel 87.1 -> 83.6 | Mixed |
| replace BRICS with finer rBRICS fragmentation (ablation) | FragFM · MOSES | Novelty: 87.1 -> 88.5 | Improved |
| replace BRICS with rBRICS fragmentation (ablation) | FragFM · ZINC250k | FCD: 0.630 -> 0.563 | Improved |
| fragment-level generation instead of atom-level graph | FragFM vs DiGress · MOSES | FCD: 1.19 -> 0.58 | Improved |
| fragment-level generation instead of atom-level graph | FragFM vs DiGress · NPGen | FCD: 2.05 -> 1.34 | Improved |
| fixed fragment vocabulary under scaffold split | FragFM · MOSES | Scaf: 10.9 vs 16.0 (Cometh-PC) | Hurt |
| reduce number of sampling steps | FragFM vs atom-based denoising models · MOSES | validity and FCD: FragFM keeps over 95% validity and FCD below 1.0 | Improved |
| add fragment-bag guidance lambda_B to classifier guidance (ablation) | FragFM · JAK2 docking score conditioning (ZINC250k) | docking score shift and validity | Improved |
| use sequence-based SAFE-GPT instead of graph models | NPGen · NPGen | novelty: Novel 73.5; NP Score KL 0.0024 | Mixed |
Closest papers by meaning
- GenMol: A Drug Discovery Generalist with Discrete Diffusion (ICML 2025 · similarity 0.89)
- Molecule Generation with Fragment Retrieval Augmentation (NeurIPS 2024 · similarity 0.88)
- Energy-Based Flow Matching for Generating 3D Molecular Structure (ICML 2025 · similarity 0.88)
- DMol: A Highly Efficient and Chemical Motif-Preserving Molecule Generation Platform (NeurIPS 2025 · similarity 0.88)
- Pretraining Generative Flow Networks with Inexpensive Rewards for Molecular Graph Generation (ICML 2025 · similarity 0.87)
- Unified all-atom molecule generation with neural fields (NeurIPS 2025 · similarity 0.87)
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.
