Research map / Graph learning / Graph self-supervised learning
Energy-Motivated Equivariant Pretraining for 3D Molecular Graphs
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 |
|---|---|---|---|
| Add the combined EFP and INP pretraining objectives | 3D-EMGP with EGNN · MD17 energy prediction (average) | MAE: 0.1191 -> 0.0876 | Helped |
| Add the combined EFP and INP pretraining objectives | 3D-EMGP with EGNN · MD17 force prediction (average) | MAE: 0.2086 -> 0.0968 | Helped |
| Add EFP alongside INP (ablation) | 3D-EMGP with EGNN · MD17 energy prediction (average) | MAE: 0.0974 -> 0.0876 | Helped |
| Add EFP alongside INP (ablation) | 3D-EMGP with EGNN · MD17 force prediction (average) | MAE: 0.1350 -> 0.0968 | Helped |
| Add INP alongside EFP (ablation) | 3D-EMGP with EGNN · MD17 energy prediction (average) | MAE: 0.0905 -> 0.0876 | Helped |
| Add INP alongside EFP (ablation) | 3D-EMGP with EGNN · MD17 force prediction (average) | MAE: 0.1193 -> 0.0968 | Helped |
| Use Riemann-Gaussian noise rather than Euclidean Gaussian noise (ablation) | 3D-EMGP with EGNN · MD17 energy prediction (average) | MAE: 0.0912 -> 0.0876 | Helped |
| Use Riemann-Gaussian noise rather than Euclidean Gaussian noise (ablation) | 3D-EMGP with EGNN · MD17 force prediction (average) | MAE: 0.1060 -> 0.0968 | Helped |
| Use energy-derived force prediction rather than direct EGNN force outputs (ablation) | 3D-EMGP with EGNN · MD17 energy prediction (average) | MAE: 0.0914 -> 0.0876 | Helped |
| Use energy-derived force prediction rather than direct EGNN force outputs (ablation) | 3D-EMGP with EGNN · MD17 force prediction (average) | MAE: 0.1267 -> 0.0968 | Helped |
| Use Riemann-Gaussian coordinate denoising rather than pairwise-distance denoising (ablation) | 3D-EMGP with EGNN · MD17 force prediction (average) | MAE: 0.1292 -> 0.0968 | Helped |
| Vary Langevin sampling steps among T=1, T=5, and T=10 (ablation) | EFP pretraining · MD17 energy prediction (average) | MAE: 0.0905; 0.0899; 0.0905 | No clear effect |
Closest papers by meaning
- 3D Denoisers Are Good 2D Teachers: Molecular Pretraining via Denoising and Cross-Modal Distillation (AAAI 2025 · similarity 0.90)
- Beyond Atoms: Enhancing Molecular Pretrained Representations with 3D Space Modeling (ICML 2025 · similarity 0.89)
- Lift Your Molecules: Molecular Graph Generation in Latent Euclidean Space (ICLR 2025 · similarity 0.89)
- Empowering Active Learning for 3D Molecular Graphs with Geometric Graph Isomorphism (NeurIPS 2024 · similarity 0.88)
- Generalist Equivariant Transformer Towards 3D Molecular Interaction Learning (ICML 2024 · similarity 0.88)
- Scalable Non-Equivariant 3D Molecule Generation via Rotational Alignment (ICML 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.