Research map / Probabilistic and scientific modeling / Molecules and proteins
Inference-time optimization for experiment-grounded protein ensemble generation
Sai Advaith Maddipatla, Anar Rzayev, Marco Pegoraro, Martin Pacesa, Paul Schanda, Ailie Marx, Sanketh Vedula, Alexander Bronstein
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 |
|---|---|---|---|
| optimize AF3 conditioning embeddings (IT-Opt) instead of coordinate-space guidance | NOE-guided AlphaFold3 · NMRDB 2K53 NOE restraints | NOE violation (%): 28.1 -> 20.6 | Improved |
| optimize AF3 conditioning embeddings (IT-Opt) instead of guidance | NOE-guided AlphaFold3 · NMRDB 2K0M NOE restraints | NOE violation (%): 11.5 -> 10.5 | Improved |
| use IT-Opt instead of guidance on a protein where guidance already fits best | NOE-guided AlphaFold3 · NMRDB 2MRW NOE restraints | NOE violation (%): 22.6 -> 24.5 | Hurt |
| add Boltzmann energy reweighting to IT-Opt | IT-Opt · NMRDB 2K0M NOE restraints | NOE violation (%): 10.5 -> 4.4 | Improved |
| add Boltzmann energy reweighting to IT-Opt | IT-Opt · NMRDB 1D3Z NOE restraints | NOE violation (%): 11.3 -> 16.4 | Hurt |
| add energy reweighting to raise effective sample size | IT-Opt · NMRDB 2K53 ProteinEBM | effective sample size: 1.002 -> 2.389 | Improved |
| add temperature annealing to energy-weighted IT-Opt (ablation) | IT-Opt (Energy) · NMRDB 6SOW NOE restraints | NOE violation (%): 32.0 -> 20.0 | Improved |
| add energy smoothing to energy-weighted IT-Opt (ablation) | IT-Opt (Energy) · NMRDB 1YEZ NOE restraints | NOE violation (%): 5.1 -> 4.5 | Improved |
| raise inverse temperature beta from 0.6 to 1.2 (ablation) | IT-Opt (Energy) · NMRDB 1YEZ NOE restraints | NOE violation (%): 4.5 -> 11.7 | Hurt |
| increase outer iterations from 10 to 20 (ablation) | IT-Opt · NMRDB 1U0P NOE restraints | NOE violation (%): 50.8 -> 44.4 | Improved |
| increase outer iterations from 20 to 30 (ablation) | IT-Opt · NMRDB 1U0P NOE restraints | NOE violation (%): 44.4 -> 43.8 | No clear effect |
| raise embedding learning rate to 1 (ablation) | IT-Opt · NMRDB 1U0P NOE restraints | NOE violation (%): 44.4 -> 48.8 | Hurt |
Closest papers by meaning
- Inverse problems with experiment-guided AlphaFold (ICML 2025 · similarity 0.92)
- AlphaFold Meets Flow Matching for Generating Protein Ensembles (ICML 2024 · similarity 0.89)
- Unlocking hidden biomolecular conformational landscapes in diffusion models at inference time (NeurIPS 2025 · similarity 0.88)
- Str2Str: A Score-based Framework for Zero-shot Protein Conformation Sampling (ICLR 2024 · similarity 0.88)
- Learning conformational ensembles of proteins based on backbone geometry (NeurIPS 2025 · similarity 0.88)
- Aligning Protein Conformation Ensemble Generation with Physical Feedback (ICML 2025 · similarity 0.88)
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.
