Thermodynamically Optimized Machine-Learned Reaction Coordinates for Hydrophobic Ligand Dissociation

Area of Science:

• Computational chemistry and biophysics
• Molecular dynamics and free energy calculations
• Machine learning applications in thermodynamics

Background:

• Ligand unbinding free energy is a crucial determinant of molecular interactions.
• Quantifying individual entropic and enthalpic contributions to unbinding free energy is challenging.
• Hydrophobic interactions play a significant role in ligand binding and dissociation.

Purpose of the Study:

• To develop and apply a deep learning framework for modeling hydrophobic ligand unbinding.
• To identify and analyze the thermodynamic contributions (entropy and enthalpy) to the unbinding free energy profile.
• To elucidate the specific mechanisms of unbinding for different sized hydrophobic ligands (methane and C60 fullerene).

Main Methods:

• Utilized a modified deep learning framework to learn a thermodynamically optimized reaction coordinate.
• Modeled all-atom simulations of methane and C60 fullerene unbinding from hydrophobic pockets in water.
• Performed feature importance analysis on the learned reaction coordinates to interpret thermodynamic forces.

Main Results:

• The free-energy barrier for both methane and C60 fullerene unbinding is primarily governed by entropic factors.
• Methane unbinding is driven by methane solvation, while fullerene unbinding involves pocket wetting followed by fullerene wetting.
• The learned reaction coordinate indicates a low direct importance of simple distance from the binding pocket.

Conclusions:

• The developed deep learning approach provides valuable thermodynamic insights into hydrophobic ligand dissociation.
• Entropy plays a dominant role in the free-energy barriers of hydrophobic ligand unbinding.
• Distinct solvation and wetting mechanisms characterize the unbinding pathways of small versus large hydrophobic ligands.