README.md

---
language:
- en
license: cc-by-4.0
size_categories:
- 1K<n<10K
task_categories:
- other
tags:
- biology
- protein-ligand-docking
configs:
- config_name: benchmark
  data_files:
  - split: test
    path:
    - astex_benchmark.csv
    - posex_self_dock_benchmark.csv
    - posex_cross_dock_benchmark.csv
---

# PoseX: AI Defeats Physics Approaches on Protein-Ligand Cross Docking

[**Paper**](https://huggingface.co/papers/2505.01700) | [**GitHub**](https://github.com/CataAI/PoseX) | [**Leaderboard**](http://dock-lab.tech)

PoseX is a comprehensive benchmark dataset designed to evaluate molecular docking algorithms for predicting protein-ligand binding poses. It includes curated datasets for both self-docking and cross-docking scenarios.

The file `posex_set.zip` contains the processed dataset for docking evaluation, while `posex_cif.zip` contains the raw CIF files from RCSB PDB. For information about creating a dataset from scratch, please refer to the data construction [code](https://github.com/CataAI/PoseX/tree/master/dataset).                                                

## Dataset Subsets
The processed dataset comprises three sub-datasets in total:
- **The Astex Diverse set**: Curated by Hartshorn et al. [1] and contains 85 cases, reprocessed to adapt to our evaluation process. 
- **The PoseX Self-Docking set**: Curated using the PDB, containing 718 cases. 
- **The PoseX Cross-Docking set**: Curated using the PDB, containing 1312 pairs. 

## Sample Usage

To generate the benchmark CSV files (containing protein sequences, ligand SMILES, and metadata) from the scripts provided in the [official repository](https://github.com/CataAI/PoseX):

```bash
# Example: Generate benchmark CSV for PoseX Self-Docking
bash ./scripts/generate_docking_benchmark.sh posex_self_dock

# Example: Generate benchmark CSV for PoseX Cross-Docking
bash ./scripts/generate_docking_benchmark.sh posex_cross_dock
```

## File Structure

For each protein-ligand complex, there is one folder named after the PDB identifier and the CCD identifier (`PDB_CCD`). 

Each folder contains the following files: 
1. `PDB_CCD_protein.pdb`: The protein structure without the ligand, without solvents, and with all cofactors. 
2. `PDB_CCD_ligands.sdf`: All instances of the ligand. 
3. `PDB_CCD_ligand.sdf`: One instance of the ligand.
4. `PDB_CCD_ligand_start_conf.sdf`: One generated molecule conformation for the ligand. 
5. `PDB_CCD.json`: Input JSON file for AlphaFold 3.
6. `PDB_CCD_ref_protein.pdb` (unique to Cross-Docking set): The reference protein structure.
7. `group_id.txt` (unique to Cross-Docking set): Group information.
8. `qtm.csv`: Contains pocket structure similarity calculated for each evaluation data point relative to the training data.

```plaintext
  posex_set/
  ├── astex_diverse_set/
  │   ├── 2BSM_BSM/
  │   │   ├── 2BSM_BSM_ligand_start_conf.sdf
  │   │   ├── 2BSM_BSM_ligand.sdf
  │   │   ├── 2BSM_BSM_ligands.sdf
  │   │   ├── 2BSM_BSM_protein.sdf
  │   │   └── 2BSM_BSM.json
  │   └── ...
  ├── posex_self_docking_set/
  │   ├── 5S9U_FAD/
  │   │   ├── 5S9U_FAD_ligand_start_conf.sdf
  │   │   ├── 5S9U_FAD_ligand.sdf
  │   │   ├── 5S9U_FAD_ligands.sdf
  │   │   ├── 5S9U_FAD_protein.pdb
  │   │   └── 5S9U_FAD.json
  │   ├── qtm.csv
  │   └── ...
  ├── posex_cross_docking_set/
  │   ├── 5SAU_1IJ/
  │   │   ├── 5SAU_1IJ_ligand_start_conf.sdf
  │   │   ├── 5SAU_1IJ_ligand.sdf
  │   │   ├── 5SAU_1IJ_ligands.sdf
  │   │   ├── 5SAU_1IJ_protein.pdb
  │   │   ├── 5SAU_1IJ_ref_protein.pdb
  │   │   ├── 5SAU_1IJ.json
  │   │   └── group_id.txt
  │   ├── qtm.csv
  └── └── ...
```

## Citation

If you use this dataset or code, please cite:

```bibtex
@inproceedings{jiang2026posex,
  title={PoseX: {AI} Defeats Physics-based Methods on Protein Ligand Cross-Docking},
  author={Yize Jiang and Xinze Li and Yuanyuan Zhang and Jin Han and Youjun Xu and Ayush Pandit and ZAIXI ZHANG and Mengdi Wang and Mengyang Wang and Chong Liu and Guang Yang and Yejin Choi and Yingzhou Lu and Wu-Jun Li and Tianfan Fu and Fang Wu and Junhong Liu},
  booktitle={The Fourteenth International Conference on Learning Representations},
  year={2026},
  url={https://openreview.net/forum?id=qqzxKudD4T}
}
```

## References

[1] Hartshorn MJ, Verdonk ML, Chessari G, Brewerton SC, Mooij WTM, Mortenson PN, et al. Diverse, high-quality test set for the validation of protein-ligand docking performance. J Med Chem. 2007 Feb 1;50(4):726–41.