Screening Fast-Mode Motion in Collective Variable Discovery for Biochemical Processes.

Area of Science:

• Computational Biology
• Biophysics
• Data Science

Background:

• Collective variables (CVs) are vital for analyzing molecular dynamics (MD) trajectories and enhancing sampling simulations.
• Existing methods like time-lagged independent component analysis (tICA) and time-lagged autoencoders (tAE) can be hindered by fast motions, complicating the extraction of slow degrees of freedom (DOFs).

Purpose of the Study:

• To develop a novel approach for accurately extracting slow degrees of freedom (DOFs) from biomolecular simulation trajectories.
• To improve the identification of collective variables (CVs) that describe essential protein dynamics and metastable states.

Main Methods:

• Integration of discrete wavelet transform (DWT) with dimensionality reduction techniques (tICA, tAE).
• DWT is used to filter out high-frequency signals corresponding to fast motions, isolating slow dynamics.
• The filtered trajectories are then analyzed using tICA and tAE to extract relevant CVs.

Main Results:

• The proposed DWT-enhanced method accurately identifies CVs representing slow DOFs, outperforming standard tICA and tAE.
• Validation on alanine dipeptide, tripeptide, and CLN025 folding demonstrates superior performance in distinguishing metastable states.
• DWT integration improves the performance of other CV-finding algorithms, including Deep-tICA, for enhanced sampling.

Conclusions:

• Discrete wavelet transform (DWT) is an effective and computationally inexpensive tool for pre-processing molecular dynamics trajectories.
• DWT significantly enhances the accuracy of collective variable extraction by effectively removing fast-motion noise.
• The proposed method offers a versatile and broadly applicable "free lunch" for improving various CV-finding algorithms in biomolecular simulations.