Research map / Probabilistic and scientific modeling / Molecules and proteins
Fractional Denoising for 3D Molecular Pre-training
Shikun Feng, Yuyan Ni, Yanyan Lan, Zhi-Ming Ma, Weiying Ma
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 |
|---|---|---|---|
| Replace coordinate denoising with hybrid-noise fractional denoising during pre-training | TorchMD-NET molecular pre-training · QM9 (12 property tasks) | MAE: Frad achieves or surpasses DP-TorchMD-NET in all 12 tasks | Improved |
| Use Frad pre-training rather than coordinate-denoising pre-training (ablation) | TorchMD-NET · QM9 (six-task aligned ablation) | MAE: μ: 0.0120 -> 0.0118; α: 0.0517 -> 0.0486; εHOMO: 17.7 -> 15.3; εLUMO: 14.3 -> 13.7; Δε: 31.8 -> 27.8; <R²>: 0.4496 -> 0. | Improved |
| Add dihedral-angle noise while keeping the coordinate-noise scale at τ=0.04 (ablation) | Coordinate denoising · QM9 (average of seven energy tasks) | Average performance (meV): 12.86 -> 11.94 | Improved |
| Increase coordinate-noise scale from τ=0.04 to τ=0.4 (ablation) | Coordinate denoising · QM9 (average of seven energy tasks) | Average performance (meV): 12.86 -> 15.26 | Hurt |
| Use traditional Noisy Nodes with τ=0.005 without decoupling task inputs (ablation) | Frad-pre-trained model · MD17 Aspirin | force MAE: 0.2141 -> do not converge | Hurt |
| Use dihedral-angle noise for Noisy Nodes (ablation) | Frad-pre-trained model · MD17 Aspirin | force MAE: 0.2141 -> 0.2096 | Improved |
| Decouple the denoising input from the downstream property-prediction input (ablation) | Noisy Nodes fine-tuning · MD17 Aspirin | force MAE: 0.2096 -> 0.2107 | Improved |
| Add fractional denoising with hybrid noise to decoupled Noisy Nodes (ablation) | Noisy Nodes fine-tuning · MD17 Aspirin | force MAE: 0.2107 -> 0.2087 | Improved |
Closest papers by meaning
- Learning 3D Anisotropic Noise Distributions Improves Molecular Force Fields (NeurIPS 2025 · similarity 0.92)
- Sliced Denoising: A Physics-Informed Molecular Pre-Training Method (ICLR 2024 · similarity 0.91)
- 3D Denoisers Are Good 2D Teachers: Molecular Pretraining via Denoising and Cross-Modal Distillation (AAAI 2025 · similarity 0.88)
- Geometric-Facilitated Denoising Diffusion Model for 3D Molecule Generation (AAAI 2024 · similarity 0.87)
- Structure-based drug design by denoising voxel grids (ICML 2024 · similarity 0.87)
- SubgDiff: A Subgraph Diffusion Model to Improve Molecular Representation Learning (NeurIPS 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.
