Scoring confidence index: statistical evaluation of ligand binding mode predictions

Area of Science:

• Computational chemistry
• Molecular modeling
• Drug discovery

Background:

• Protein-ligand docking generates numerous binding poses, necessitating reliable scoring functions for accurate prediction.
• Current scoring functions exhibit target-dependent performance in predicting native binding modes.
• Assessing the confidence of docking predictions is crucial for effective virtual screening.

Purpose of the Study:

• To develop a statistical approach for measuring the confidence of protein-ligand scoring performance.
• To introduce a novel metric, the scoring confidence index (SCI), for evaluating binding mode prediction reliability.
• To enhance the accuracy of virtual screening by providing a measure of scoring reliability.

Main Methods:

• Developed a correlation-based score (CBScore) using the correlation coefficient between docking scores and distances to the native pose.
• Combined CBScore with score range to create the scoring confidence index (SCI).
• Validated SCI on 50 protein-ligand complexes using AffiScore, DrugScore, and X-Score.

Main Results:

• High SCI values indicate smooth energy landscapes and reliable binding mode predictions.
• Low SCI values suggest rugged energy landscapes, leading to difficult and unreliable predictions.
• Binding mode predictions were three times more reliable for complexes with high SCI values.

Conclusions:

• The scoring confidence index provides a robust measure of scoring performance in protein-ligand docking.
• SCI is a valuable tool for virtual screening, improving the reliability of binding mode predictions.
• This approach aids in identifying promising drug candidates by enhancing the accuracy of molecular docking assessments.