Research map / Probabilistic and scientific modeling / Dynamical systems
Learning Collective Variables from BioEmu with Time-Lagged Generation
Seonghyun Park, Kiyoung Seong, Soojung Yang, Rafael Gomez-Bombarelli, Sungsoo Ahn
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 time-lagged conditioning (target x_{t+tau} instead of x_t) (ablation) | BioEmu time-lagged CV · Chignolin steered MD | target hit percentage: 81.3 -> 100.0 | Improved |
| add time-lagged conditioning (ablation) | BioEmu time-lagged CV · Chignolin steered MD | C-alpha RMSD (A): 1.57 -> 1.20 | Improved |
| add time-lagged conditioning (ablation) | BioEmu time-lagged CV · Chignolin OPES | |dF_ref - dF| error: 2.10 -> 0.52 | Improved |
| add time-lagged conditioning (ablation) | BioEmu time-lagged CV · Chignolin OPES | PMF MAE: 1.41 -> 3.07 | Hurt |
| freeze BioEmu weights instead of unfreezing (ablation) | BioEmu time-lagged CV · Chignolin steered MD | C-alpha RMSD (A): 1.62 -> 1.20 | Improved |
| freeze BioEmu weights instead of unfreezing (ablation) | BioEmu time-lagged CV · Chignolin OPES | |dF_ref - dF| error: 1.22 -> 0.52 | Improved |
| increase encoder size from 48K to 1.27M parameters (ablation) | BioEmu time-lagged CV · Chignolin steered MD | C-alpha RMSD (A): 1.20 -> 1.50 | No clear effect |
| increase encoder size from 48K to 196K parameters (ablation) | BioEmu time-lagged CV · Chignolin steered MD | target hit percentage: 100.0 -> 93.8 | No clear effect |
| also condition the pair representation (single, pair) (ablation) | BioEmu time-lagged CV · Chignolin steered MD | target hit percentage: 100.0 -> 68.8 | Hurt |
| condition the pair representation instead of single (ablation) | BioEmu time-lagged CV · Chignolin steered MD | target hit percentage: 100.0 -> 56.3 | Hurt |
| replace DeepTICA with time-lagged BioEmu CV | DeepTICA · Chignolin steered MD | target hit percentage: 37.5 -> 100.0 | Improved |
| replace DeepTICA with time-lagged BioEmu CV | DeepTICA · BBA steered MD | target hit percentage: 18.8 -> 93.8 | Improved |
Closest papers by meaning
- Consistent Sampling and Simulation: Molecular Dynamics with Energy-Based Diffusion Models (NeurIPS 2025 · similarity 0.87)
- Simultaneous Modeling of Protein Conformation and Dynamics via Autoregression (NeurIPS 2025 · similarity 0.87)
- TEMPO: Temporal Multi-scale Autoregressive Generation of Protein Conformational Ensembles (NeurIPS 2025 · similarity 0.86)
- Transition Path Sampling with Improved Off-Policy Training of Diffusion Path Samplers (ICLR 2025 · similarity 0.86)
- Action-Minimization Meets Generative Modeling: Efficient Transition Path Sampling with the Onsager-Machlup Functional (ICML 2025 · similarity 0.86)
- Unlocking hidden biomolecular conformational landscapes in diffusion models at inference time (NeurIPS 2025 · similarity 0.85)
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.
