Research map / Probabilistic and scientific modeling / Molecules and proteins
Structure-based 3D molecule generation
Generates drug-like molecules in protein pockets and extends scaffolds with 3D generative models.
107 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2019–2026) · 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.
30 papers in 2023–24 and 57 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 6.1% to 6.1%.
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
- 2019 · A Model to Search for Synthesizable Molecules (NeurIPS 2019 · 71 citations)
- 2020 · Multi-Objective Molecule Generation using Interpretable Substructures (ICML 2020 · 94 citations)
- 2021 · MARS: Markov Molecular Sampling for Multi-objective Drug Discovery (ICLR 2021 · 29 citations)
- 2022 · Pocket2Mol: Efficient Molecular Sampling Based on 3D Protein Pockets (ICML 2022 · 78 citations)
- 2023 · DiffDock: Diffusion Steps, Twists, and Turns for Molecular Docking (ICLR 2023 · 362 citations)
- 2024 · GxVAEs: Two Joint VAEs Generate Hit Molecules from Gene Expression Profiles (AAAI 2024 · 12 citations)
- 2025 · SynFlowNet: Design of Diverse and Novel Molecules with Synthesis Constraints (ICLR 2025 · 12 citations)
- 2026 · PepBenchmark: A Standardized Benchmark for Peptide Machine Learning (ICLR 2026 · 2 citations)
Earliest papers here
- A Model to Search for Synthesizable Molecules (NeurIPS 2019 · 71 citations)
- Molecular Hypergraph Grammar with Its Application to Molecular Optimization (ICML 2019 · 50 citations)
- Multi-Objective Molecule Generation using Interpretable Substructures (ICML 2020 · 94 citations)
What has been tried
68 papers here report 690 measured changes (a component added, removed or swapped, compared with a baseline): helped 67%, no clear effect 5%, hurt 24%, 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 |
|---|---|---|---|---|
| increase maximum trajectory length from 3 to 4 | sEH molecule generation · AiZynth retrosynthesis success | 62% -> 40% | Hurt | SynFlowNet: Design of Diverse and Novel Molecules with Synthesis Constraints (ICLR 2025 · 12 citations) |
| Use noise-reduced parameter-space sampling instead of original sampling | CrossDocked test, same checkpoint · Vina Score | -5.42 -> -6.51 | Helped | MolCRAFT: Structure-Based Drug Design in Continuous Parameter Space (ICML 2024 · 11 citations) |
| Increase building block library size | KOR pocket-specific optimization · Docking score | Helped | Generative Flows on Synthetic Pathway for Drug Design (ICLR 2025 · 5 citations) | |
| use exact-energy preference optimization instead of ranked-pair DPO | CrossDocked2020 test · Avg. Vina Score | -6.81 -> -7.07 | Helped | Aligning Target-Aware Molecule Diffusion Models with Exact Energy Optimization (NeurIPS 2024 · 5 citations) |
| use entropy ranking instead of uniform decoding | BPTI · JS-TIC | 0.428 -> 0.402 | Helped | Structure Language Models for Protein Conformation Generation (ICLR 2025 · 4 citations) |
| add product context at each sampling step | USPTO-50k validation · top-5 exact-match accuracy | 75.21 -> 79.44 | Helped | RetroBridge: Modeling Retrosynthesis with Markov Bridges (ICLR 2024 · 4 citations) |
| Increase noise level from σ=0.9 to σ=1.0 | CrossDocked2020 test · VinaScore Avg. | -6.94 -> -6.63 | Hurt | Structure-based drug design by denoising voxel grids (ICML 2024 · 4 citations) |
| replace geometric interaction with VFN's operation | CATH4.3 Full · recovery | 0.53 -> 0.52 | Hurt | UniIF: Unified Molecule Inverse Folding (NeurIPS 2024 · 4 citations) |
| Apply joint gradient guidance to coordinates and atom types rather than single-modality guidance | CrossDocked2020 molecule optimization · molecular properties and affinity | Helped | Empower Structure-Based Molecule Optimization with Gradient Guided Bayesian Flow Networks (ICML 2025 · 3 citations) | |
| Add guidance injection to CByG | Kinase selectivity test set · Success rate | 62.6 % -> 78.3 % | Helped | Controllable 3D Molecular Generation for Structure-Based Drug Design Through Bayesian Flow Networks and Gradient Integration (NeurIPS 2025 · 3 citations) |
| Replace tokenization + self-attention and cross-attention with binning + MLP and concatenation | NIST · Accuracy | 49.0% -> 4.30% | Hurt | MADGEN: Mass-Spec attends to De Novo Molecular generation (ICLR 2025 · 3 citations) |
| Set α=1, removing posterior-based negative-sample reweighting | Contrastive learning objective · conditional mutual information lower bound | Mixed | Removing Biases from Molecular Representations via Information Maximization (ICLR 2024 · 3 citations) |
What reviewers will expect
Datasets and baselines named in the 41 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- CrossDocked2020 · 7 of 41 papers
- ZINC250k · 3 of 41 papers
- PoseBusters · 2 of 41 papers
- ZINC · 2 of 41 papers
- MassSpecGym · 2 of 41 papers
- ChEMBL · 2 of 41 papers
Baselines
- DecompDiff · 10 of 41 papers
- Pocket2Mol · 9 of 41 papers
- TargetDiff · 9 of 41 papers
- REINVENT · 8 of 41 papers
- MolCRAFT · 8 of 41 papers
- AR · 6 of 41 papers
- DiffSBDD · 4 of 41 papers
- RGA · 3 of 41 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 87 papers; from OpenAlex and the papers' affiliations).
- University of Toronto · 7 papers
- Microsoft · 6 papers
- Massachusetts Institute of Technology · 6 papers
- Tsinghua University · 5 papers
- University of Science and Technology of China · 4 papers
- Vector Institute · 4 papers
- Shanghai Institute of Materia Medica · 3 papers
- University of Montreal · 3 papers
- Stanford University · 3 papers
- Renmin University of China · 2 papers
Use it
Download BibTeX of 38 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
- Diffusion for molecular conformations (67 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).