Approximating dynamic proximity with a hybrid geometry energy-based kernel for diffusion maps.

Area of Science:

• Computational chemistry
• Biophysics
• Data science

Background:

• Diffusion maps are dimensionality reduction techniques.
• Their accuracy depends on the kernel's dynamic measure.
• Root-mean-square deviation (RMSD) is commonly used but has limitations.

Purpose of the Study:

• To develop a more accurate diffusion map embedding for molecular systems.
• To improve the description of conformational transitions by incorporating both geometry and energy.

Main Methods:

• A novel hybrid geometry-energy kernel was developed for diffusion maps.
• This kernel considers both RMSD and energy differences between conformations.
• The method was applied to analyze molecular dynamics trajectories of a β-hairpin and Trp-cage.

Main Results:

• The hybrid kernel improved free energy landscape characterization for the β-hairpin compared to RMSD-only.
• A new correlation measure revealed enhanced insights into cluster dynamics.
• Analysis of Trp-cage folding provided a unified view of its mechanism.

Conclusions:

• The hybrid geometry-energy kernel enhances diffusion map accuracy for molecular dynamics.
• This approach offers a more robust method for analyzing molecular system dynamics and thermodynamics.
• The hybrid criterion has potential as a general dynamic measure for broader applications.