Uni-Dock-Benchmarks

The Uni-Dock-Benchmarks repository provides a comprehensive collection of datasets for benchmarking the Uni-Dock docking system's performance and accuracy. The datasets include prepared structures and input files for both Uni-Dock V1 and V2 for benchmarks.

Data

Benchmark data within the repository is categorized into two primary sections:

• molecular_docking
• virtual_screening

Molecular Docking Benchmarks

Under the molecular_docking directory, you will find several well-known benchmark datasets:

• Astex: Hartshorn, M. J., Verdonk, M. L., Chessari, G., Brewerton, S. C., Mooij, W. T., Mortenson, P. N., & Murray, C. W. (2007). Diverse, high-quality test set for the validation of protein ligand docking performance. Journal of medicinal chemistry, 50(4), 726-741.
• CASF2016: Su, M., Yang, Q., Du, Y., Feng, G., Liu, Z., Li, Y., & Wang, R. (2018). Comparative assessment of scoring functions: the CASF-2016 update. Journal of chemical information and modeling, 59(2), 895-913.
• PoseBuster: Buttenschoen, M., Morris, G. M., & Deane, C. M. (2024). PoseBusters: AI-based docking methods fail to generate physically valid poses or generalise to novel sequences. Chemical Science.

We performed the following preparation steps for the proteins and ligands in the datasets.

• After obtaining the protein structures from the RCSB database based on the PDB code, we retained the crystal waters that affect the binding mode and completed missing protein side chains and lost hydrogen atoms.
• For ligands, we searched the RCSB database for the isomer SMILES corresponding to the PDB code and determined the correct protonation state according to the receptor pocket environment. Then, we generated 3D conformations for each ligand.

After excluding systems for covalent ligand bindings, problematic binding mechanisms and those with large natural products or polypeptide ligands, 69 systems from Astex, 271 systems from CASF-2016 and 396 systems from PoseBuster were used as benchmarks.

The correctness of protein side chain structure and hydrogen bond networks have crucial impact on ligand docking, and hence the structure preparation for both protein and ligand determines the difficultness of producing correct ligand docking poses. We use our internal tools to prepare the initial structures of receptor and ligands so that we can obtain better docking results. In addition, we also integrated the open-sourced version of structure preparation algorithmd for Uni-Dock V2 into the unified protocol in the Uni-Dock V2 github repository.

We prepare the receptor structure in two versions, protein with co-crystalized water version and protein only verison, to test the overall effect of the presence of water opn ligand docking experiments.

The directory structure for each dataset is as follows:

<DataSetName> <PDB_ID> <PDB_ID>_ligand.sdf # Ligand co-crystal structure processed in SDF format <PDB_ID>_protein_water_cleaned.pdb # Prepared receptor structrue with protein and crystalized water in PDB format <PDB_ID>_protein_cleaned.pdb # Prepared receptor structrue with only protein in PDB format ligand_prepared.sdf # Reprepared ligand 3D conformation used in docking test in SDF format unidock1_protein # Folder for input files of Uni-Dock V1, with protein only in the receptor structure ligand_prepared_torsion_tree.sdf # Prepared ligand structure with torsion tree information used in Uni-Dock V1 input in SDF format receptor.pdbqt # Prepared receptor structure used in Uni-Dock V1 input in PDBQT format unidock1_protein_water # Folder for input files of Uni-Dock V1, with protein and water in the receptor structure ligand_prepared_torsion_tree.sdf # Prepared ligand structure with torsion tree information used in Uni-Dock V1 input in SDF format receptor.pdbqt # Prepared receptor structure used in Uni-Dock V1 input in PDBQT format unidock2_protein # Folder for input files of Uni-Dock V2, with protein only in the receptor structure <PDB_ID>_unidock2.json # Integrated JSON input file for Uni-Dock V2 docking engine receptor_parameterized.dms # Prepared and parameterized receptor structure in DMS format unidock2_protein_water # Folder for input files of Uni-Dock V2, with protein and water in the receptor structure <PDB_ID>_unidock2.json # Integrated JSON input file for Uni-Dock V2 docking engine receptor_parameterized.dms # Prepared and parameterized receptor structure in DMS format pdb_center.csv # CSV file recording the protein pocket center with respect to the <PDB_ID> for each system

Virtual Screening Benchmarks

Under the virtual_screening directory, you will find several meticulously selected benchmark datasets:

• D4: Lyu, J., Wang, S., Balius, T. E., Singh, I., Levit, A., Moroz, Y. S., ... & Irwin, J. J. (2019). Ultra-large library docking for discovering new chemotypes. Nature, 566(7743), 224-229.
• GBA: Tran-Nguyen, V. K., Jacquemard, C., & Rognan, D. (2020). LIT-PCBA: an unbiased data set for machine learning and virtual screening. Journal of chemical information and modeling, 60(9), 4263-4273.
• NSP3: Schuller, M., Correy, G. J., Gahbauer, S., Fearon, D., Wu, T., Daz, R. E., ... & Ahel, I. (2021). Fragment binding to the Nsp3 macrodomain of SARS-CoV-2 identified through crystallographic screening and computational docking. Science advances, 7(16), eabf8711.
• PPARG: Tran-Nguyen, V. K., Jacquemard, C., & Rognan, D. (2020). LIT-PCBA: an unbiased data set for machine learning and virtual screening. Journal of chemical information and modeling, 60(9), 4263-4273.
• sigma2: Alon, A., Lyu, J., Braz, J. M., Tummino, T. A., Craik, V., OMeara, M. J., ... & Kruse, A. C. (2021). Structures of the 2 receptor enable docking for bioactive ligand discovery. Nature, 600(7890), 759-764.

The following table summarizes the statistics of the datasets:

| Dataset | PDB ID | NActives | NInactives | N_Total | |----|----|----|----|----| | D4 | 5WIU | 226 | 598 | 824 | | GBA | 5LVX | 286 | 458,205 | 458,491 | | NSP3 | 5RS7 | 65 | 3,515 | 3,580 | | PPARG | 5Y2T | 29 | 7,292 | 7,321 | | sigma2 | 7M94 | 228 | 596 | 824 |

The directory structure for each dataset is as follows:

<DataSetName> docking_grid.json # JSON file recording the protein pocket center and the box sizes <PDB_ID>_receptor.pdb # Original unprocessed receptor structure in PDB format <PDB_ID>_protein_cleaned.pdb # Prepared receptor structure with only protein in PDB format actives_cleaned.sdf # Preprocessed and cleaned active molecules in SDF format actives.sdf # Active molecules in SDF format inactives_cleaned.sdf # Preprocessed and cleaned inactive molecules in SDF format inactives.sdf # Inactive molecules in SDF format unidock1_protein # Folder for input files of Uni-Dock V1, with protein only in the receptor structure actives_prepared_torsion_tree.sdf # Prepared active molecule structure with torsion tree information used in Uni-Dock V1 input in SDF format inactives_prepared_torsion_tree.sdf # Prepared inactive molecule structure with torsion tree information used in Uni-Dock V1 input in SDF format receptor.pdbqt # Prepared receptor structure used in Uni-Dock V1 input in PDBQT format unidock2_protein # Folder for input files of Uni-Dock V2, with protein only in the receptor structure actives_unidock2.json # Integrated JSON input file of active molecules for Uni-Dock V2 docking engine inactives_unidock2.json # Integrated JSON input file of inactive molecules for Uni-Dock V2 docking engine receptor_parameterized.dms # Prepared and parameterized receptor structure in DMS format

Important Note Due to the substantial number of inactive molecules in the GBA dataset, the directory contains several large files that exceed GitHub's size limits. These files have been moved to cloud storage. To obtain the complete GBA directory, please run the following command in your terminal: sh ./getGBA.sh

Scripts

Scripts are provided to run tests directly on Uni-Dock2 executable binary. Simply run molecular docking test on Uni-Dock2 engine binary: sh python scripts/run_tests.py --version 2 --bin ud2 --type molecular_docking --nowater --device 1 --savedir my_res --seed 121

Please read scripts/run_tests.py for complete argument documentation. Key parameters include:

• --version <1|2> For example, you can use --version 1 if you are testing on Uni-Dock 1https://github.com/dptech-corp/Uni-Dock

NOTE Specifying --version 2 automatically loads the file scripts/ud2.yaml as default configuration.

• --bin <PATH> Path of the tested executable binary
• --type <molecular_docking|virtual_screening>
• --nowater Select receptor without water. (Default: uses water-containing receptor)

ATTENTION: Since docking searches are always incomplete, we recommend: 1. Repeating tests with multiple random seeds (--seed <INTEGER>) 2. Average results across repetitions.