Research map / Probabilistic and scientific modeling / Molecules and proteins
Diffusion for molecular conformations
Generates 3D conformers using gradient fields, geometric diffusion and neural dynamics.
67 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2020–2026) · an approach in Molecules and proteins, keyword cluster “molecules · drug · chemical”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
19 papers in 2023–24 and 37 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 3.9% to 4.0%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2020 · TorsionNet: A Reinforcement Learning Approach to Sequential Conformer Search (NeurIPS 2020 · 8 citations)
- 2021 · GeoMol: Torsional Geometric Generation of Molecular 3D Conformer Ensembles (NeurIPS 2021 · 52 citations)
- 2022 · GeoDiff: A Geometric Diffusion Model for Molecular Conformation Generation (ICLR 2022 · 179 citations)
- 2023 · Pre-training via Denoising for Molecular Property Prediction (ICLR 2023 · 30 citations)
- 2024 · Transferable Boltzmann Generators (NeurIPS 2024 · 11 citations)
- 2025 · Stiefel Flow Matching for Moment-Constrained Structure Elucidation (ICLR 2025 · 2 citations)
- 2026 · Protein Language Model Embeddings Improve Generalization of Implicit Transfer Operators (ICML 2026 · 1 citation)
Earliest papers here
- TorsionNet: A Reinforcement Learning Approach to Sequential Conformer Search (NeurIPS 2020 · 8 citations)
- GeoMol: Torsional Geometric Generation of Molecular 3D Conformer Ensembles (NeurIPS 2021 · 52 citations)
- Learning Gradient Fields for Molecular Conformation Generation (ICML 2021 · 39 citations)
What has been tried
40 papers here report 405 measured changes (a component added, removed or swapped, compared with a baseline): helped 65%, no clear effect 5%, hurt 25%, mixed 4%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add amino-acid and sequence-position embeddings to the atom encoding | Unseen dipeptides · ESS | 0.58±1.04\% -> 15.29±9.27\% | Helped | Transferable Boltzmann Generators (NeurIPS 2024 · 11 citations) |
| remove atom features | GEOM-QM9 · Precision AMR mean | 0.103 -> 0.187 | Hurt | Swallowing the Bitter Pill: Simplified Scalable Conformer Generation (ICML 2024 · 8 citations) |
| replace discretised basis charge with continuous basis charge | QM9 · molecule stability | 90.87 -> 89.94 | Hurt | Unified Generative Modeling of 3D Molecules with Bayesian Flow Networks (ICLR 2024 · 3 citations) |
| add stochastic sampling | GEOM-DRUGS · Precision Coverage mean | 74.38 -> 75.19 | Helped | ET-Flow: Equivariant Flow-Matching for Molecular Conformer Generation (NeurIPS 2024 · 3 citations) |
| add equivariant optimal transport | QM9 test · % < RMSD 0.25 Å | 15.17±0.31 -> 13.99±0.30 | Hurt | Stiefel Flow Matching for Moment-Constrained Structure Elucidation (ICLR 2025 · 2 citations) |
| Select training conformations with GOLF data-collecting scheme vs training on MMFF-optimized conformations (rdkit) | nablaDFT test · pct_T / pct_success | pct_T 98.8 vs 93.0; success 77.3 vs 35.4 (GOLF-10k vs rdkit) | Helped | Gradual Optimization Learning for Conformational Energy Minimization (ICLR 2024 · 2 citations) |
| Replace Q-MPNN with PointNet++ in the router | ATLAS test (All) · IoU | 0.7028±0.0390 -> 0.6656 | Hurt | Generalized Implicit Neural Representations for Dynamic Molecular Surface Modeling (AAAI 2025 · 1 citation) |
| replace E3WAE with FactorVAE-style E3fVAE | GEOM-Drugs Asphericity · MSE | 0.095 -> 0.143 | Hurt | Learning Disentangled Equivariant Representation for Explicitly Controllable 3D Molecule Generation (AAAI 2025 · 0 citations) |
| Add MAE loss to RMSD-only training | Molecule3D (random) validation · C-RMSD | 0.716 -> 0.662 | Helped | WGFormer: An SE(3)-Transformer Driven by Wasserstein Gradient Flows for Molecular Ground-State Conformation Prediction (ICML 2025 · 0 citations) |
| Use JAMUN instead of full diffusion sampling with the same architecture | Timewarp 2AA · JSD-Backbone Torsions | 0.1501 vs 0.1363 (3149 samples); 0.0496 vs 0.0460 (200000 samples) | Mixed | JAMUN: Bridging Smoothed Molecular Dynamics and Score-Based Learning for Conformational Ensemble Generation (NeurIPS 2025 · 0 citations) |
| Add OrbDiff_WN to DiffCSP | Perov-5 · Match | 52.02 -> 52.39 | Helped | Rao-Blackwell Gradient Estimators for Equivariant Denoising Diffusion (NeurIPS 2025 · 0 citations) |
| Use subtractive rather than additive anisotropic correction | MD17 Aspirin · force MAE | 0.1128 -> 0.1118 | Helped | Learning 3D Anisotropic Noise Distributions Improves Molecular Force Fields (NeurIPS 2025 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 22 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- GEOM-Drugs · 6 of 22 papers
- QM9 · 6 of 22 papers
- ATLAS · 2 of 22 papers
- GEOM-QM9 · 2 of 22 papers
Baselines
- EDM · 2 of 22 papers
- GeoDiff · 2 of 22 papers
- FoldFlow-OT · 2 of 22 papers
- GeoMol · 2 of 22 papers
- MCF · 2 of 22 papers
- ETFlow · 2 of 22 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 56 papers; from OpenAlex and the papers' affiliations).
- Tsinghua University · 8 papers
- Stanford University · 3 papers
- Renmin University of China · 3 papers
- University of Oxford · 3 papers
- Academy of Mathematics and Systems Science · 3 papers
- Westlake University · 2 papers
- Google DeepMind · 2 papers
- National Research University Higher School of Economics · 2 papers
- University of Chinese Academy of Sciences · 2 papers
- Mila · 2 papers
Use it
Download BibTeX of 36 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Structure-based 3D molecule generation (107 papers)
- Machine-learned interatomic potentials (74 papers)
- Graph models for retrosynthesis (10 papers)
- Structure-based molecule design with diffusion (48 papers)
Elsewhere in the topic
- Generative protein structure design (142 papers)
- Prediction-powered and conformal inference (129 papers)
- Statistical analysis of ML practice (122 papers)
- Neuro-symbolic probabilistic layers (120 papers)
- Sparse graphical and tensor models (97 papers)
- Flow matching generative modeling (97 papers)
- Machine learning for physical simulation (96 papers)
- Multimodal trajectory prediction (95 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).