Protein-Ligand Complex Structures with Filtered Ligands

Comprising approximately 36,000 unique pairs of protein sequences and ligand SMILES strings, along with the 3D coordinates of their complexes from the Protein Data Bank (PDB). Ligands are filtered to have at least 3 atoms, a molecular weight of 100 Da or more, and exclude the 280 most common PDB ligands. It was created by author jglaser and last updated in October 2022.

Use Cases

• Train a model to predict protein-ligand complex coordinates from protein sequences and ligand SMILES strings.
• Analyze structural patterns in complexes involving ligands with molecular weight >= 100 Da.
• Filter and study non-common ligands by excluding the 280 most frequent PDB additives like PEG and ADP.
• Use tokenized SMILES strings (via the Schwaller regex) for sequence-based molecular property prediction.

Strengths

• Approximately 36,000 unique protein-ligand pairs provide a substantial sample size.
• Ligand selection criteria (>=3 atoms, MW >= 100 Da, excludes 280 common ligands) ensure a focused, non-trivial chemical space.
• Includes 3D structural coordinates for the complexes, enabling geometric analysis.

Limitations

• The dataset excludes common ligands and additives, which may limit applicability for studying prevalent binding motifs.
• The specific source PDB entries and their temporal range are not detailed, potentially affecting reproducibility.
• The total number of rows, columns, and file formats are unknown, complicating assessment of data structure and volume.

Provenance

Source

Protein Data Bank (PDB)

Collection Method

Extracted and filtered protein-ligand complexes from PDB based on specified ligand criteria.

Freshness

Last updated on 2022-10-13.