Research map / Probabilistic and scientific modeling / Molecules and proteins
SynCoGen: Synthesizable 3D Molecule Generation via Joint Reaction and Coordinate Modeling
Andrei Rekesh, Miruna Cretu, Dmytro Shevchuk, Pietro Lio, Robert A. Batey, Mike Tyers, Michał Koziarski, Cheng-Hao Liu
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 |
|---|---|---|---|
| use SynCoGen reaction-graph plus coordinate co-generation instead of atom and coordinate generation (SemlaFlow) | 3D molecule generation · SynSpace de novo generation | Syntheseus solve rate (%): 36 -> 72 | Improved |
| use SynCoGen instead of SemlaFlow retrained on SynSpace (ablation) | 3D molecule generation · SynSpace de novo generation | RDKit validity (%): 72.0 -> 96.7 | Improved |
| use SynCoGen instead of SemlaFlow retrained on SynSpace (ablation) | 3D molecule generation · SynSpace de novo generation | PoseBusters validity (%): 60.3 -> 87.2 | Improved |
| use SynCoGen instead of SemlaFlow on diversity | 3D molecule generation · SynSpace de novo generation | diversity: 0.85 -> 0.78 | Hurt |
| use SynCoGen instead of SemlaFlow on novelty | 3D molecule generation · SynSpace de novo generation | novelty (%): 99.6 -> 93.9 | Hurt |
| add chemistry-sensitive graph constraints and self-conditioning (ablation) | SynCoGen · SynSpace de novo generation | generation quality | Improved |
| add inference annealing on coordinates (ablation) | SynCoGen · sampling | performance | Improved |
| use SynCoGen instead of DiffLinker for fragment linking | fragment linking · fragment linking (3 targets) | retrosynthesis solve rate (%): 0 -> 58-79 | Improved |
| use SynCoGen instead of ShEPhERD, CGFlow-ZS and SynFormer for pharmacophore-conditioned generation | pharmacophore conditioning · 10 targets (PDB and LIT-PCBA) | best docking score wins: best on 8/10 targets (best sample), 7/10 (median) | Improved |
| train on larger SynSpace-L vocabulary (ablation) | SynCoGen · SynSpace-L de novo generation | validity and synthesizability | No clear effect |
Closest papers by meaning
- Compositional Flows for 3D Molecule and Synthesis Pathway Co-design (ICML 2025 · similarity 0.90)
- Controllable 3D Molecular Generation for Structure-Based Drug Design Through Bayesian Flow Networks and Gradient Integration (NeurIPS 2025 · similarity 0.89)
- BindGPT: A Scalable Framework for 3D Molecular Design via Language Modeling and Reinforcement Learning (AAAI 2025 · similarity 0.88)
- SynFlowNet: Design of Diverse and Novel Molecules with Synthesis Constraints (ICLR 2025 · similarity 0.88)
- FlashMol: High-Quality Molecule Generation in as Few as Four Steps (NeurIPS 2026 · similarity 0.88)
- Projecting Molecules into Synthesizable Chemical Spaces (ICML 2024 · 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.
