Improved ligand-protein binding affinity predictions using multiple binding modes.

Area of Science:

• Computational Chemistry
• Drug Discovery
• Biochemistry

Background:

• Accurate prediction of ligand-protein binding affinity is a significant challenge in drug development, particularly during lead optimization.
• Traditional docking and scoring functions often yield unsatisfactory results for similar binders.
• Cytochrome P450 enzymes present difficulties for binding affinity predictions due to large, flexible active sites.

Purpose of the Study:

• To develop an improved computational method for predicting ligand-protein binding affinity.
• To address the challenge of insufficient sampling in free energy calculations for flexible binding sites.
• To reduce the dependency on accurate initial pose selection in docking studies.

Main Methods:

• An iterative scheme employing multiple independent molecular dynamics simulations was introduced.
• Weighted ensemble averages were calculated from simulations for use in the linear interaction energy method.
• The method was applied to predict binding affinities for 12 compounds against Cytochrome P450 2C9.

Main Results:

• The proposed iterative scheme automatically weights various ligand poses, reducing the criticality of initial pose selection.
• The method accounts for multiple binding modes contributing to overall affinity or compounds occupying different poses.
• A root-mean-square error of 2.9 kJ/mol was achieved for binding affinity predictions.

Conclusions:

• The developed iterative molecular dynamics approach enhances the accuracy of ligand-protein binding affinity predictions.
• This method is particularly beneficial for challenging targets like Cytochrome P450s with flexible binding sites.
• The improved prediction accuracy aids in more effective lead optimization during drug development.