T38DrugDB: Medicinally Relevant Compounds in the ZINC20

T38DrugDB - Medicinally Relevant Compounds in the ZINC20 The database lives at:- [Zenodo] (Large data files) - [github] (Executables and README)- [CSBJ] (Original Publication) Overview T38DrugDB provides: - 34,345,597 compounds in PDBQT and MOL2 formats - Filtered dataset based on LogP, molecular weight, and medicinal chemistry guidelines - Random subsampling tools for flexible dataset creation Usage This directory contains a subset of the ZINC20 database filtered for physicochemical and structural properties observed in medicinal compounds. The purpose of this database is to aid drug discovery, especially concerning virtual screening via autodock-GPU. However, any other chemoinformatics tool for drug discovery that relies on MOL2 or PDBQT files may benefit from this subset. The main component is the PDBQT_fitlered subdirectory that contains >34,000,000 compounds in PDBQT format deposited in a .zip archive. The idea is to subsample randomly from this database via fetch_N_random_PDBQTs.py (make sure all required modules are installed). The .zip format was chosen to keep the database portable; however, this causes performance bottlenecks when trying to access the compounds. Using fetch_N_random_PDBQTs to subsample around 100,000 compounds may take 10 minutes, depending on the hardware. Therefore, tinyDB was created, which contains a tiny subset with fewer .zip archives. The Python file fetch_N_random_PDBQTs.py can also be found there with identical functionality. Any testing should be conducted in this subdirectory for faster results. The file example.sh should extract 100 random PDBQT compounds into tempPDBQT/ and showcase the functionality of this database quickly (several seconds). Quickstart Test functionality with a Python script:Download only tinyDB/ cd tinyDBpython fetchNrandomPDBQTs.py -i tinyT38DrugDB.csv -N 100 -d ./tempPDBQT/ Or use the provided example script to extract 100 random PDBQT compounds into `tempPDBQT/` within seconds. bash ./example.sh Use the Jupyter notebook for more detailed exploration: jupyter notebook fetchNrandomPDBQTs.ipynb Compounds can also be extracted manually via their ZINC ID and corresponding tranche. For PDBQT format: unzip -p PDBQT_filtered/HFABMN.zip ZINC000006949878.pdbqt > test.pdbqt For MOL2 format unzip -p MOL2_filtered/FFADMN.zip ZINC000620995125.mol2 > test.mol2 Directory Structure T38DrugDB/ ├── README.md # THIS file ├── T38drugDB.csv # Compound Table (34M+ rows, 3 cols) ├── fetch_N_random_PDBQTs.py # PDBQT sampling script ├── fetch_N_random_MOL2s.py # MOL2 sampling script ├── PDBQT_filtered/ # Main database (PDBQT format 35 GB) │ └── [trancheID].zip # Compressed compound batches ├── MOL2_filtered/ # Main database (MOL2 format 75 GB) │ └── [trancheID].zip # Compressed compound batches └── tinyDB/ # TEST SUBSET (1.2 GB) ├── example.sh # Quick test script ├── fetch_N_random_PDBQTs.py # see above ├── fetch_N_random_MOL2s.py # see above ├── fetch_N_random_PDBQTs.ipynb # interactive subsampling ├── tinyPDBQT_filtered/ # see above PDBQT_filtered/ │ └── [trancheID].zip # see above └── tinyMOL2_filtered/ # see above MOL2_filtered/ └── [trancheID].zip # see above Directory Description tinyDB/ - lightweight database for testing and quick downloading - contains only 1.2 GB instead of 110 GB of Data T38drugDB.csv - list of all SMILES and ZINCIDs contained in T38DrugDB.zip - auxiliary file for subsampling compounds from the database - 34,345,597 lines fetchNrandom_PDBQTs.py - Python executable to subsample the database randomly - the file T38drugDB.csv is used for subsampling - PDBQT files are extracted from the zip files in PDBQT_filtered - PDBQT files are extracted into a directory specified as a command line argument to the python script (default: see python fetch_N_random_PDBQTs.py --help) - It can take up to 10 minutes for a subsample ranging in the hundreds of thousands fetchNrandomMOL2s.py - same as fetchNrandomPDBQTs.py but for MOL2 format MOL2_filtered/ - subdirectory with 75 GB of data - contains downloaded compounds from the ZINC20 database - filtered by LogP and Molecular Weight via the ZINC20 webpage - also filtered by the Handbook of Medicinal Chemistry - compounds are individually stored in zip archives - compounds are stored as plain text MOL2 format in batch files: .zip PDBQTfiltered/ - subdirectory with 35 GB of data - contains downloaded compounds from the ZINC20 database - filtered by LogP and Molecular Weight via the ZINC20 webpage - also filtered by the Handbook of Medicinal Chemistry - compounds are individually stored in zip archives - compounds are stored as plain text PDBQT format in batch files: .zip * tinyDB/fetchNrandomPDBQTs.ipynb - jupyter notebook for interactive testing - This file subsamples the database randomly - the file T38drugDB.csv is used for subsampling - PDBQT files are extracted from the zip files in PDBQT_filtered - PDBQT files are extracted into a directory specified as an argument in the parser.parse_args() function example.sh - Quick demonstration script Optimization Tips Use tinyDB/ for development and testing Consider writing custom parallel processing scripts for extensive extractions Filter Criteria Applied to ZINC20 Database: | property | range (including) | |-------------------------|-------------------| | LogP | 1 to 3 | | Mass | 250 to 450 Da | | NOCount | 2 to 9 | | NHOHCount | 0 to 3 | | TPSA | 19 to 70 A^2 | | NumRotatableBonds | 1 to 8 | | NumAromaticCarbocycles | 0 to 2 | | NumAromaticHeterocycles | 0 to 1 | | NumAromaticRings | 1 to 3 | | FractionCSP3 | 0.2 to 0.6 | Citation If you use T38DrugDB in your research, please cite: Luis Vollmers, Shu-Yu Chen, Martin Zacharias. In Silico Analysis of Potential Stabilizer Binding Sites at Protein–RNA Interfaces. Comput Struct Biotechnol J. 2026;35:0016. DOI:10.34133/csbj.0016 License This work is licensed under a Creative Commons Attribution 4.0 International License. See creativecommons.org/licenses/by/4.0/ for further information. Contact For questions, issues, or contributions: - [e-mail] - [e-mail] - Publication Link: https://doi.org/10.34133/csbj.0016