RCSB Ligand Retrieval: 72,388 Protein-Ligand Complexes for Drug Discovery

RCSB Protein Data Bank provides 72,388 commercially redistributable protein-ligand complexes. The dataset, created by HemanthKari, is built for retrieval-augmented structure-based drug discovery and was initially released in April 2026.

Use Cases

• Train retrieval models for drug candidate screening based on protein pocket embeddings.
• Benchmark ligand similarity search algorithms based on Morgan fingerprint embeddings.
• Develop affinity prediction models based on staff-curated BindingDB affinity data.
• Augment molecular docking pipelines with pre-computed structural complexes.

Strengths

• Contains 72,388 protein-ligand complexes, providing substantial scale for model training.
• All components are commercially redistributable, facilitating commercial application.
• Includes multiple data modalities: FAISS index, ESM-2 embeddings, Morgan fingerprints, HDF5 coordinates, and curated affinities.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for specific tasks.

Provenance

Source

RCSB Protein Data Bank

Collection Method

Retrieval dataset built from publicly available structural data.

Time Range

Initial release date is 2026-04 15.

Freshness

Last updated 2026-04-23 22:47:31; freshness should be verified.

Geography

Global coverage of structural biology data.