Combined Free-Energy Calculation and Machine Learning Methods for Understanding Ligand Unbinding Kinetics

Area of Science:

• Computational chemistry
• Drug discovery
• Molecular dynamics

Background:

• Drug residence time determination is crucial but experimentally costly.
• Current methods for measuring kinetic rate constants are time-consuming and expensive.
• Computational approaches are needed to accelerate the drug discovery process.

Purpose of the Study:

• To computationally determine drug residence times.
• To develop a novel algorithm for identifying molecular design objectives based on ligand unbinding kinetics.
• To enable the design of drugs with improved residence times.

Main Methods:

• Enhanced sampling technique to predict free-energy profiles of ligand unbinding.
• Biased molecular dynamics (MD) simulations to identify important internal coordinates (ICs).
• Finite-temperature string simulations to calculate the free-energy barrier for unbinding.
• Supervised machine learning (ML) approach using unbiased trajectories to identify key ligand-protein interactions.

Main Results:

• Accurate prediction of free-energy barriers for ligand unbinding.
• Identification of key ligand-protein interactions driving the unbinding process.
• Successful application to cyclin-dependent kinase 2 (CDK2) inhibitors, with results comparable to experimental data.
• Highlighting distal ligand interactions for residence time optimization.

Conclusions:

• The developed method provides a novel computational tool for determining unbinding rates.
• It identifies key structural features for targeted drug design and optimization of residence times.
• This approach can accelerate the discovery of new drugs, particularly kinase inhibitors for cancer treatment.