LigandExplorer: An Automated Tool for Ligand Extraction from PDB Structures.

Area of Science:

• Structural biology
• Computational chemistry
• Drug discovery

Background:

• Protein-ligand complex structural data is vital for drug design.
• Existing databases often have inconsistent annotations and require extensive processing for large-scale modeling.

Purpose of the Study:

• To develop an automated pipeline, LigandExplorer, for processing biomolecular complex structures.
• To identify, extract, and categorize ligands (covalent and noncovalent) for standardized downstream use.

Main Methods:

• Utilizes residue-level graphs from atomic coordinates, independent of metadata.
• Integrates LightGBM models for ligand classification (peptides, nucleic acids, phospholipids, carbohydrates, organics, ions) and interaction relevance assessment.
• Designed as a rerunnable pipeline for continuous updates.

Main Results:

• Achieved 98.38% structural agreement on the PDBbind v2020 refined set.
• Successfully processed 97.52% of complexes in the PepBDB dataset.
• Identified discrepancies primarily due to upstream data inconsistencies.

Conclusions:

• LigandExplorer automates data cleaning and standardization for biomolecular complex structures.
• Enables the creation of rapidly refreshed, machine-ready datasets for computational modeling and small-molecule design.
• Mitigates data-cleaning burdens in drug discovery research.