Molecular Free Energies, Rates, and Mechanisms from Data-Efficient Path Sampling Simulations

Area of Science:

• Computational Chemistry
• Biophysics
• Machine Learning

Background:

• Molecular dynamics simulations are crucial for studying molecular thermodynamics and kinetics.
• Standard simulations often fail to capture the necessary time scales for complex events.
• Transition path sampling (TPS) enhances sampling but can yield non-equilibrium distributions.

Purpose of the Study:

• To develop an algorithm for approximating equilibrium path ensembles from machine learning-guided path sampling data.
• To enable efficient calculation of molecular mechanisms, free energies, and rates for rare events.
• To address the limitations of non-Boltzmann-distributed trajectories in ML-enhanced TPS.

Main Methods:

• Integration of machine learning with transition path sampling.
• Development of a novel algorithm to reconstruct equilibrium path ensembles.
• Application of the method to the folding of the mini-protein chignolin.

Main Results:

• The algorithm efficiently samples rare molecular events at moderate computational cost.
• It provides accurate approximations of mechanism, free energy, and rates.
• Demonstrated successful application to protein folding dynamics.

Conclusions:

• The developed algorithm offers a data-efficient and straightforward approach to analyze complex molecular systems.
• It overcomes key limitations of existing path sampling and machine learning integration methods.
• Opens new avenues for studying challenging molecular events in biophysics and chemistry.