Wavelet-Enhanced Data-Driven Collective Variables for Efficient Sampling of Protein Folding Landscapes

Area of Science:

• Computational chemistry and biophysics
• Molecular dynamics simulations
• Enhanced sampling techniques

Background:

• Collective variables (CVs) are crucial for enhanced sampling methods in biomolecular simulations.
• Selecting appropriate geometric CVs is challenging and system-dependent for complex free-energy landscapes.
• Data-driven CVs are susceptible to fast fluctuations, obscuring slow degrees of freedom (DOFs).

Purpose of the Study:

• To develop a robust method for identifying biologically relevant slow modes in molecular simulations.
• To overcome the limitations of existing data-driven CVs in handling fast stochastic fluctuations.
• To enhance the efficiency of enhanced sampling methods for complex biomolecular systems.

Main Methods:

• Utilized discrete wavelet transform for feature extraction to filter fast-mode motions from molecular dynamics (MD) trajectories.
• Applied dimensionality reduction to the filtered descriptors.
• Constructed data-driven low-dimensional CVs using the refined feature subset.

Main Results:

• The proposed strategy successfully identified slow DOFs with high fidelity.
• CVs derived from this method facilitated rapid transitions between folded and unfolded states.
• Demonstrated applicability to both chignolin and the complex BBA protein system.

Conclusions:

• The feature extraction strategy effectively filters noise and reveals slow dynamics in MD simulations.
• This approach significantly accelerates the exploration of protein folding landscapes.
• The method provides a reliable way to construct accurate CVs for enhanced sampling across diverse systems.