Research map / Image generation / Autoregressive and latent generation
Geometric Latent Diffusion Models for 3D Molecule Generation
Minkai Xu, Alexander S Powers, Ron O. Dror, Stefano Ermon, Jure Leskovec
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 equivariant point-structured latent features rather than invariant-only latents (ablation) | GeoLDM · QM9 | validity (%): 83.6 -> 93.8 | Improved |
| Use equivariant point-structured latent features rather than invariant-only latents (ablation) | GeoLDM · GEOM-DRUG | validity (%): 97.2 -> 99.3 | Improved |
| Add random-rotation augmentation while using invariant-only latent modeling (ablation) | GraphLDM · QM9 | validity (%): 83.6 -> 90.5 | Improved |
| Add random-rotation augmentation while using invariant-only latent modeling (ablation) | GraphLDM · GEOM-DRUG | validity (%): 97.2 -> 98.0 | Improved |
| Use ES-reg early stopping instead of KL-reg (ablation) | GeoLDM (k=1) · QM9 | validity (%): 83.7 -> 93.8 | Improved |
| Reduce latent invariant dimension from k=16 to k=1 (ablation) | GeoLDM with ES-reg · QM9 | validity (%): 92.4 -> 93.8 | Improved |
| Reduce latent invariant dimension from k=8 to k=1 (ablation) | GeoLDM with ES-reg · QM9 | validity (%): 92.1 -> 93.8 | Improved |
| Reduce latent invariant dimension from k=4 to k=1 (ablation) | GeoLDM with ES-reg · QM9 | validity (%): 92.6 -> 93.8 | Improved |
| Model generation in equivariant latent space instead of atomic feature space | GeoLDM vs EDM · QM9 conditional generation | MAE (Bohr^3): 2.76 -> 2.37 | Improved |
| Model generation in equivariant latent space instead of atomic feature space | GeoLDM vs EDM · QM9 conditional generation | MAE (meV), Δε: 655 -> 587 | Improved |
| Model generation in equivariant latent space instead of atomic feature space | GeoLDM vs EDM · QM9 conditional generation | MAE (meV), εHOMO: 356 -> 340 | Improved |
| Model generation in equivariant latent space instead of atomic feature space | GeoLDM vs EDM · QM9 conditional generation | MAE (meV), εLUMO: 584 -> 522 | Improved |
Closest papers by meaning
- Towards Unified and Lossless Latent Space for 3D Molecular Latent Diffusion Modeling (NeurIPS 2025 · similarity 0.92)
- Geometric-Facilitated Denoising Diffusion Model for 3D Molecule Generation (AAAI 2024 · similarity 0.92)
- Enhancing Generalizability in Molecular Conformation Generation with METRIZATION-Informed Geometric Diffusion Pretraining (AAAI 2025 · similarity 0.91)
- Equivariant Neural Diffusion for Molecule Generation (NeurIPS 2024 · similarity 0.91)
- Geometric Trajectory Diffusion Models (NeurIPS 2024 · similarity 0.90)
- Unified Generative Modeling of 3D Molecules with Bayesian Flow Networks (ICLR 2024 · similarity 0.90)
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.
