Large-scale Docking Scores for 250 Million Molecules Across Eight Therapeutic Targets

Two large-scale virtual screening datasets for benchmarking machine learning methods in early-phase drug discovery. The datasets, provided by Andreas Luttens of Uppsala University, contain canonical SMILES, compound identifiers, and docking scores for approximately 15.5 million 'Rule-of-Four' molecules and approximately 235 million 'lead-like' molecules docked against eight different biological targets.

Use Cases

• Benchmarking machine learning models for docking score prediction based on molecular SMILES strings.
• Training models to prioritize subsets of ultralarge chemical libraries based on provided docking scores.
• Analyzing the relationship between molecular libraries ('Rule-of-Four' vs 'lead-like') and docking performance across different protein targets.

Strengths

• Contains docking scores for approximately 250.5 million molecules combined across two distinct chemical libraries.
• Covers eight different biological targets representing different protein classes and binding sites.
• Includes canonical SMILES and compound identifiers, which likely facilitate molecular structure analysis.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.
• Row count for individual datasets is implied but not explicitly confirmed in a structured metadata field.

Provenance

Source

Uppsala University (Carlsson Lab), shared via paperswithcode and GitHub.

Collection Method

Virtual screening campaigns docking chemical libraries against biological targets.