BINDKIN

BINDKIN benchmarks the impact of point mutations on kinase–ligand binding affinity to assess predictive methods and quantify mutation-induced changes in binding.

Key Features:

• Dataset composition: Contains 23 pairs of wild-type and mutant kinase–ligand complexes with associated experimental binding affinity data (IC50, Kd, Ki).
• Biological scope: Focuses on protein kinases, including mutations within or near the ATP-binding pocket that affect phosphorylation activity and can confer drug resistance.
• Predictor benchmarking: Evaluates six web-based affinity predictors: DSX-ONLINE, KDEEP, HADDOCK2.2, PDBePISA, Pose&Rank, and PRODIGY-LIG.
• Analytical comparison: Compares raw affinity predictions against experimental data and additionally analyzes the direction of change (improvement or worsening of binding) caused by mutations.
• Performance highlights: Reports that DSX-ONLINE achieved Pearson’s R = 0.97 using kinetic Ki data, with correlation decreasing to 0.45 when homology models were used instead of crystal structures.
• Structural model assessment: Explicitly compares prediction accuracy using crystal structures versus homology models to evaluate model-dependent effects.

Scientific Applications:

• Predictor evaluation: Benchmarking and comparison of scoring functions and web-based affinity prediction servers.
• Mutation impact studies: Quantifying how point mutations in kinase ATP-binding pockets alter ligand binding and contribute to drug resistance.
• Modeling validation: Assessing the effect of using crystal structures versus homology models on the accuracy of affinity predictions.
• Drug discovery research: Informing development and optimization of kinase-targeted inhibitors and studies of resistance mechanisms.

Methodology:

Assembled a dataset of 23 wild-type/mutant kinase–ligand complexes with IC50, Kd, and Ki values; benchmarked six predictors (DSX-ONLINE, KDEEP, HADDOCK2.2, PDBePISA, Pose&Rank, PRODIGY-LIG); compared raw predictions versus direction-of-change analysis; computed Pearson’s R correlations; and compared results using crystal structures and homology models.