Research map / Image generation / Autoregressive and latent generation
MolDiff: Addressing the Atom-Bond Inconsistency Problem in 3D Molecule Diffusion 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 |
|---|---|---|---|
| jointly diffuse atoms and bonds with bond-first noise schedule and bond predictor guidance (MolDiff vs EDM) | 3D molecule diffusion · GEOM-Drug | success rate: 0.371 -> 0.993 | Helped |
| predict bonds after atom generation with a lookup table instead of diffusing bonds (ablation) | MolDiff without bond diffusion · GEOM-Drug | success rate: 0.993 -> 0.278 | Hurt |
| include bonds during diffusion (vs post-hoc bond prediction with NN predictor) (ablation) | MolDiff · GEOM-Drug | success rate: 0.993 vs 0.935 (NN predictor) and 0.961 (Open Babel) | Helped |
| continuous embedding of atom and bond types instead of discrete categorical diffusion (ablation) | MolDiff · GEOM-Drug | success rate: 0.993 -> 0.808 (0.907 with scaling) | Hurt |
| add bond-length MSE loss (ablation) | MolDiff · GEOM-Drug | success rate: 0.993 -> 0.958 | Hurt |
| remove bond predictor guidance (ablation) | MolDiff · GEOM-Drug | RMSD: 0.939 -> 1.027 | Hurt |
| add bond predictor guidance (ablation) | MolDiff · GEOM-Drug | JS bond angles / dihedral angles: dihedral 0.202 -> 0.162 | Helped |
| remove bond-first noise schedule (ablation) | MolDiff · GEOM-Drug | JS bond lengths: 0.365 -> 0.461 | Hurt |
| add bond-first noise schedule (ablation) | MolDiff · GEOM-Drug | JS number of rings: 0.176 -> 0.062 (no schedule vs full) | Helped |
| remove bond-first noise schedule (ablation) | MolDiff · GEOM-Drug | success rate / validity: success 0.993 -> 0.993 | No clear effect |
| add bond predictor guidance and bond-first schedule on JS basic bond types (ablation) | MolDiff · GEOM-Drug | JS basic bond types: 0.093 vs No Gui. 0.054 and No Sche. 0.039 | Mixed |
| MolDiff vs EDM on QM9 | 3D molecule diffusion · QM9 | success rate: 0.918 -> 0.968 | Helped |
Closest papers by meaning
- Geometric-Facilitated Denoising Diffusion Model for 3D Molecule Generation (AAAI 2024 · similarity 0.90)
- All-atom Diffusion Transformers: Unified generative modelling of molecules and materials (ICML 2025 · similarity 0.89)
- Navigating the Design Space of Equivariant Diffusion-Based Generative Models for De Novo 3D Molecule Generation (ICLR 2024 · similarity 0.88)
- Towards Unified and Lossless Latent Space for 3D Molecular Latent Diffusion Modeling (NeurIPS 2025 · similarity 0.87)
- 3D Molecule Generation from Rigid Motifs via $\mathrm{SE}(3)$ Flows (NeurIPS 2026 · similarity 0.87)
- Enhancing Generalizability in Molecular Conformation Generation with METRIZATION-Informed Geometric Diffusion Pretraining (AAAI 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.