Lost in Projection? Gaussian Filtering Recovers Hidden Conformational States.

Area of Science:

• Computational chemistry
• Biophysics
• Statistical mechanics

Background:

• Molecular dynamics (MD) simulations are crucial for understanding molecular behavior.
• Dimensionality reduction is commonly used to interpret complex MD data by projecting onto collective variables.
• Projection artifacts can distort free energy landscapes, leading to inaccurate identification of conformational states.

Purpose of the Study:

• To investigate a novel method for improving the accuracy of free energy landscape reconstruction from MD data.
• To demonstrate how Gaussian low-pass filtering can mitigate projection artifacts.
• To enhance the identification and characterization of metastable states in biomolecular systems.

Main Methods:

• Applied Gaussian low-pass filtering to high-dimensional feature trajectories from MD simulations.
• Analyzed a two-dimensional toy model to validate the filtering approach.
• Tested the method on an all-atom folding trajectory of the HP35 protein.

Main Results:

• Gaussian filtering successfully restored the underlying free energy landscape in the toy model, recovering previously hidden states.
• The number of microstates for the HP35 folding trajectory increased significantly (by an order of magnitude).
• Metastable states identified using the filtered data were longer-lived and structurally better defined.

Conclusions:

• Gaussian low-pass filtering is an effective technique to correct projection artifacts in MD data analysis.
• This method significantly improves the characterization of conformational states and their dynamics.
• The approach offers a more accurate representation of biomolecular free energy landscapes, aiding in the study of protein folding and other complex processes.