Combining Machine Learning and Enhanced Sampling Techniques for Efficient and Accurate Calculation of Absolute...

Area of Science:

• Computational Chemistry
• Molecular Dynamics
• Drug Discovery

Background:

• Calculating absolute binding free energies is crucial but challenging in computational chemistry.
• Existing methods often face limitations in accuracy, computational cost, or applicability to complex systems.

Purpose of the Study:

• To evaluate recently developed metadynamics-based methods for binding free energy calculations.
• To develop and assess a novel combination of metadynamics with Hamiltonian replica-exchange.
• To investigate the accuracy, efficiency, and applicability of these methods for diverse ligands.

Main Methods:

• Metadynamics simulations utilizing a funnel-shaped restraint.
• Combination of metadynamics with machine learning-optimized pathlike variables.
• Integration of metadynamics with the SWISH Hamiltonian replica-exchange algorithm.
• Testing on 18 chemically diverse ligands binding to human soluble epoxide hydrolase.

Main Results:

• Metadynamics with funnel-shaped restraints accurately and affordably calculates absolute binding free energy for small fragments.
• The combination with ML-optimized variables or SWISH yields reasonably accurate results for more complex ligands.
• The metadynamics-SWISH combination provides a good balance of computational cost and speed.
• This combined approach offers insights into the role of water in ligand binding mechanisms.

Conclusions:

• Metadynamics-based methods, particularly when combined with SWISH, present a computationally efficient and accurate approach for absolute binding free energy calculations.
• The developed methods are suitable for a range of ligand complexities and provide valuable mechanistic insights.
• This work advances the capabilities of computational drug discovery by improving the prediction of ligand-target interactions.