Unfolding Hidden Barriers by Active Enhanced Sampling.

Area of Science:

• Computational Chemistry and Physics
• Machine Learning in Molecular Dynamics
• Enhanced Sampling Techniques

Background:

• Enhanced sampling algorithms using collective variables (CVs) are vital for exploring complex potential energy landscapes.
• Degeneracy in microscopic configurations perpendicular to CVs can obscure hidden barriers, reducing sampling efficiency.
• Existing methods struggle to address these degeneracies systematically during simulations.

Purpose of the Study:

• To introduce a machine learning-based approach for systematically improving collective variable (CV) based enhanced sampling.
• To address and iteratively lift degeneracies in sampling arising from orthogonal degrees of freedom.
• To enhance the efficiency and accuracy of exploring complex molecular systems.

Main Methods:

• Development of an active learning scheme integrating a deep neural network-based CV learner with a CV-based enhanced sampler.
• Iterative refinement of CVs by identifying least informative sampling regions.
• Establishment of a positive feedback loop between the CV learner and the enhanced sampler.

Main Results:

• Demonstration that systematic machine learning CVs can iteratively lift sampling degeneracies on the fly.
• The active enhanced sampling algorithm effectively identifies under-sampled regions.
• Preservation of global kinetic characteristics through enhanced sample completeness and CV quality.

Conclusions:

• Machine learning-driven collective variables offer a powerful strategy to overcome limitations in enhanced sampling.
• The proposed active learning scheme significantly improves the efficiency of exploring complex systems.
• This approach provides a robust method for enhancing both sample quality and CV accuracy in simulations.