Weighted Distance Functions Improve Analysis of High-Dimensional Data: Application to Molecular Dynamics Simulations

Area of Science:

• Computational chemistry
• Biophysics
• Data science

Background:

• Data mining requires effective feature representation and distance metrics.
• High-dimensional data, like molecular dynamics (MD) simulations, often contain irrelevant features that obscure relevant information.
• Unsupervised learning on MD data needs distance measures accounting for feature relevance.

Purpose of the Study:

• To develop a data-driven approach for weighting features in MD simulations.
• To emphasize slow degrees of freedom that indicate metastable states.
• To enhance unsupervised learning for analyzing complex molecular systems.

Main Methods:

• Proposing a method to globally or locally weight simulation features based on effective rates.
• Coupling feature weighting with unsupervised learning protocols (clustering, dimensionality reduction).
• Applying the approach to analyze miniprotein and protein dynamics.

Main Results:

• Successfully unmasked slow side chain dynamics within a miniprotein's native state.
• Revealed additional metastable conformations in a protein system.
• Demonstrated the effectiveness of feature weighting in enhancing data analysis.

Conclusions:

• Feature weighting based on effective rates is a powerful tool for MD data analysis.
• This method improves the ability of unsupervised learning to identify key molecular dynamics and conformational states.
• The approach is versatile and can be integrated with various clustering and dimensionality reduction algorithms.