Research map / Image generation / Autoregressive and latent generation
Align Your Structures: Generating Trajectories with Structure Pretraining for Molecular Dynamics
Aniketh Iyengar, Jiaqi Han, Pengwei Sun, Mingjian Jiang, Jianwen Xie, Stefano Ermon
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 |
|---|---|---|---|
| add conformer structure pretraining (ablation) | EGInterpolator · QM9 unconditional generation | Bond length JSD (mean): 0.583 -> 0.210 | Improved |
| add conformer structure pretraining (ablation) | EGInterpolator · Drugs forward simulation | Bond length JSD (mean): 0.386 -> 0.142 | Improved |
| add conformer structure pretraining (ablation) | EGInterpolator · QM9 unconditional generation | TICA_0,1 JSD (mean): 0.680 -> 0.636 | Improved |
| use cascaded interpolation architecture instead of a residual temporal stack (ablation) | EGInterpolator · Drugs forward simulation | Torsion JSD (mean): 0.414 -> 0.377 | Improved |
| use cascaded interpolation architecture instead of a residual temporal stack (ablation) | EGInterpolator · Drugs forward simulation | Bond length JSD (mean): 0.330 -> 0.142 | Improved |
| use Casc variant instead of Simple | EGInterpolator · QM9 unconditional generation | Bond angle JSD (mean): 0.357 -> 0.305 | Improved |
| use Casc variant instead of Simple | EGInterpolator · Drugs forward simulation | Bond length JSD (mean): 0.258 -> 0.142 | Improved |
| condition the first block on a ground-truth frame (forward simulation vs unconditional) (ablation) | EGInterpolator-Casc · Drugs | Bond length JSD (mean): 0.195 -> 0.142 | Improved |
| train unconditionally without first frame, compared with GeoTDM forward | EGInterpolator-Casc-U · Drugs | Bond length JSD (mean): 0.643 -> 0.195 | Improved |
| joint training on QM9 and Drugs | EGInterpolator-Both · QM9 unconditional generation | Bond length JSD (mean): 0.210 -> 0.168 | Improved |
| select checkpoint 539 over 99 (precision-oriented) | BasicES · QM9 conformers | COV-P mean (%): 55.23 -> 58.12 | Improved |
| select checkpoint 539 over 99 (precision-oriented) | BasicES · QM9 conformers | COV-R mean (%): 90.18 -> 87.62 | Hurt |
Closest papers by meaning
- Generative Modeling of Molecular Dynamics Trajectories (NeurIPS 2024 · similarity 0.94)
- Atomic Trajectory Modeling with State Space Models for Biomolecular Dynamics (NeurIPS 2026 · similarity 0.90)
- Energy-Based Flow Matching for Generating 3D Molecular Structure (ICML 2025 · similarity 0.90)
- Simultaneous Modeling of Protein Conformation and Dynamics via Autoregression (NeurIPS 2025 · similarity 0.89)
- Swallowing the Bitter Pill: Simplified Scalable Conformer Generation (ICML 2024 · similarity 0.89)
- Action-Minimization Meets Generative Modeling: Efficient Transition Path Sampling with the Onsager-Machlup Functional (ICML 2025 · similarity 0.89)
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.
