PDBbind

PDBbind compiles a curated dataset of protein–ligand complexes with experimentally determined dissociation rate constants (k_off) to support quantitative structure–kinetics relationship (QSKR) modeling.

Key Features:

• Extensive Data Compilation: The PDBbind-koff-2020 dataset comprises 680 unique protein–ligand complexes characterized by experimentally determined k_off values curated from PDBbind entries.
• Structural Diversity: The dataset covers 155 distinct protein types with k_off values spanning nearly ten orders of magnitude.
• Structural Data: Three-dimensional structures are retrieved from the Protein Data Bank (PDB) or modeled using appropriate templates.
• Modeling and Prediction: A random forest (RF) model was developed using protein–ligand atom pair descriptors to predict k_off, with both experimentally determined and modeled structures used as training samples.

Scientific Applications:

• QSKR model development: Provides paired structural data and dissociation rate constants for training and evaluating quantitative structure–kinetics relationship models.
• Model benchmarking: Supplies a random forest baseline predictor for comparison against other predictive QSKR approaches.
• Structure-based drug design: Enables analysis of binding/unbinding kinetics to inform the kinetic aspects of drug–target interaction studies.

Methodology:

Structures were retrieved from the Protein Data Bank or modeled using templates, and a random forest (RF) model was developed using protein–ligand atom pair descriptors with both experimentally determined and modeled structures as training samples.