A High-Quality Data Set of Protein-Ligand Binding Interactions Via Comparative Complex Structure Modeling

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• High-quality protein-ligand complex structures are crucial for understanding noncovalent binding and structure-based drug design.
• Experimentally determined structures are limited compared to the vast chemical space, hindering comprehensive analysis.

Purpose of the Study:

• To address the scarcity of experimental protein-ligand complex structures.
• To construct a large-scale, high-quality dataset of modeled protein-ligand complexes with experimental binding affinity data.
• To provide a resource for investigating protein-ligand interactions, structure-activity relationships, and evaluating computational methods.

Main Methods:

• Comparative complex structure modeling was employed to construct the BindingNet dataset.
• The dataset was curated to contain 69,816 modeled protein-ligand complexes with experimental binding affinity data.
• The dataset was designed for visual inspection, interpretation of structure-activity relationships, and evaluation of machine learning models.

Main Results:

• BindingNet provides valuable insights into protein-ligand interactions and structure-activity relationships.
• Machine learning models trained on BindingNet demonstrated reduced bias compared to models trained on PDBbind.
• The study discussed strategies for improving BindingNet and its use in benchmarking molecular docking and binding free energy calculations.

Conclusions:

• BindingNet complements existing datasets like PDBbind, offering a sufficient and unbiased resource for protein-ligand binding studies.
• The dataset facilitates the evaluation and improvement of computational tools for drug design and binding affinity prediction.
• BindingNet is freely available, promoting further research in structural biology and computational chemistry.