ZINC-Curated Drug-Like Molecules in SMILES Format

427 million drug-like molecules in SMILES format were curated from the ZINC-22 database using a multi-stage filtration pipeline. The dataset prioritizes pharmaceutical relevance through stringent physicochemical and medicinal chemistry rules. It was created by SZU-ADDG and last updated in February 2026.

Use Cases

• Train generative models on SMILES strings for de novo molecular design in drug discovery.
• Apply quantitative structure-activity relationship (QSAR) modeling using physicochemical properties derived from SMILES data.
• Benchmark molecular property prediction models on a large-scale, curated collection of drug-like molecules.
• Analyze structural diversity and scaffold distributions within the SMILES-formatted molecular library.

Strengths

• Contains approximately 427 million drug-like molecules, providing a massive scale for training.
• Constructed via a rigorous multi-stage filtration pipeline from ZINC-22 to enforce quality and pharmaceutical relevance.
• Prioritizes medicinal chemistry rules and structural validity, focusing on drug-like chemical space.

Limitations

• The specific columns, features, and file formats are not detailed, limiting understanding of the data structure.
• The sample data is unavailable, preventing preliminary inspection of data quality and format.
• The license, organization details, and exact size are unknown, which may affect usability and attribution.

Provenance

Source

Curated subset of the ZINC-22 database.

Collection Method

Constructed using a rigorous multi-stage filtration pipeline enforcing physicochemical constraints and medicinal chemistry rules.

Freshness

Last updated on 2026-02-22.