Research map / 3D vision / 3D generation
Molecule Joint Auto-Encoding: Trajectory Pretraining with 2D and 3D Diffusion
weitao Du, Jiujiu Chen, Xuecang Zhang, Zhi-Ming Ma, Shengchao 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 |
|---|---|---|---|
| pretrain with MoleculeJAE instead of random init | SchNet · QM9 U0 | MAE: 13.27 -> 10.70 | Improved |
| pretrain with MoleculeJAE instead of GeoSSL-DDM | SchNet · QM9 energy tasks | MAE: 10.96 -> 10.70 | Improved |
| MoleculeJAE vs GeoSSL-DDM on R^2 | SchNet · QM9 R^2 | MAE: 0.141 vs 0.168 GeoSSL; best 0.116 | Hurt |
| pretrain with MoleculeJAE instead of random init | SchNet · MD17 Uracil | MAE: 0.773 -> 0.463 | Improved |
| pretrain with MoleculeJAE instead of random init | SchNet · MD17 Aspirin | MAE: 1.203 -> 1.289 | Hurt |
| pretrain with MoleculeJAE instead of GeoSSL-DDM | SchNet · MD17 Aspirin | MAE: 1.107 -> 1.289 | Hurt |
| remove contrastive term (lambda2 0.01 -> 0) (ablation) | MoleculeJAE · QM9 U0 | MAE: 11.42 -> 10.70 | Improved |
| raise contrastive weight lambda2 from 0.01 to 1 (ablation) | MoleculeJAE · QM9 U0 | MAE: 10.70 -> 13.25 | Hurt |
| remove contrastive term (lambda2 0.01 -> 0) (ablation) | MoleculeJAE · MD17 Uracil | MAE: 0.590 -> 0.463 | Improved |
| remove contrastive term on MD17 Naphthalene (ablation) | MoleculeJAE · MD17 Naphthalene | MAE: 0.482 vs 0.498 | Mixed |
| raise contrastive weight lambda2 from 0.01 to 1 (ablation) | MoleculeJAE · MD17 Uracil | MAE: 0.463 -> 0.809 | Hurt |
| fine-tune MoleculeJAE with discrete diffusion vs MiDi | MiDi backbone · GEOM-DRUGS generation molecule stability | percent: 69.2 -> 84.5 | Improved |
Closest papers by meaning
- Enhancing Molecular Representation Learning Through the Combination of 3D and 2D Graph Machine Learning (Student Abstract) (AAAI 2025 · similarity 0.87)
- Beyond Atoms: Enhancing Molecular Pretrained Representations with 3D Space Modeling (ICML 2025 · similarity 0.87)
- Learning Over Molecular Conformer Ensembles: Datasets and Benchmarks (ICLR 2024 · similarity 0.86)
- 3D Denoisers Are Good 2D Teachers: Molecular Pretraining via Denoising and Cross-Modal Distillation (AAAI 2025 · similarity 0.86)
- Multimodal Molecular Pretraining via Modality Blending (ICLR 2024 · similarity 0.86)
- NExT-Mol: 3D Diffusion Meets 1D Language Modeling for 3D Molecule Generation (ICLR 2025 · similarity 0.86)
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.
