Research map / Image generation / Autoregressive and latent generation
Torsional Diffusion for Molecular Conformer Generation
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 |
|---|---|---|---|
| diffuse only over torsion angles (torsional diffusion) instead of Euclidean diffusion (GeoDiff) | conformer generator · GEOM-DRUGS | recall AMR median: 0.809 -> 0.565 | Helped |
| diffuse over torsion angles instead of Euclidean, precision | conformer generator · GEOM-DRUGS | precision AMR median: 1.090 -> 0.729 | Helped |
| torsional diffusion vs OMEGA commercial software | conformer generator · GEOM-DRUGS | recall coverage mean: 53.4 -> 72.7 | Helped |
| increase denoising steps from 5 to 20 (ablation) | Torsional Diffusion · GEOM-DRUGS | recall AMR median: 0.685 -> 0.565 | Helped |
| increase denoising steps from 10 to 20 (ablation) | Torsional Diffusion · GEOM-DRUGS | recall AMR median: 0.580 -> 0.565 | No clear effect |
| use 20 steps instead of 5000 steps of GeoDiff | Torsional Diffusion · GEOM-DRUGS | runtime (core-secs per conformer): 305 -> 4.90 | Helped |
| torsional diffusion vs RDKit on runtime | Torsional Diffusion · GEOM-DRUGS | runtime: 0.10 -> 4.90 | Hurt |
| torsional diffusion vs baselines on dipole moment | Torsional Diffusion · ensemble properties | median error dipole: 0.34 (GeoMol) vs 0.35 | No clear effect |
| torsional diffusion for ensemble energy and minimum energy | Torsional Diffusion · ensemble properties | median error energy E (kcal/mol): 0.31 (GeoDiff) -> 0.22 | Helped |
| torsional Boltzmann generator vs annealed importance sampling | Torsional BG · Boltzmann sampling ESS | effective sample size at 1000K: 6.72 (AIS 100 steps) -> 11.42 | Helped |
| increase torsional BG steps from 5 to 20 at test time (trained with 5-step resampler) (ablation) | Torsional BG · Boltzmann sampling ESS | effective sample size at 1000K: 7.28 -> 11.42 | Helped |
| train on ground-truth local structures instead of conformer matching (ablation) | Torsional Diffusion · GEOM-DRUGS | RMSD | Hurt |
Closest papers by meaning
- DiSCO: Diffusion Schrödinger Bridge for Molecular Conformer Optimization (AAAI 2024 · similarity 0.89)
- Swallowing the Bitter Pill: Simplified Scalable Conformer Generation (ICML 2024 · similarity 0.88)
- Equivariant Blurring Diffusion for Hierarchical Molecular Conformer Generation (NeurIPS 2024 · similarity 0.87)
- Enhancing Generalizability in Molecular Conformation Generation with METRIZATION-Informed Geometric Diffusion Pretraining (AAAI 2025 · similarity 0.87)
- Efficient Molecular Conformer Generation with SO(3)-Averaged Flow Matching and Reflow (ICML 2025 · similarity 0.86)
- Unlocking hidden biomolecular conformational landscapes in diffusion models at inference time (NeurIPS 2025 · similarity 0.86)
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.