A Minimum Variance Clustering Approach Produces Robust and Interpretable Coarse-Grained Models.

Area of Science:

• Computational Biology
• Biophysics
• Statistical Mechanics

Background:

• Markov state models (MSMs) are essential for analyzing molecular dynamics (MD) data, particularly protein folding simulations.
• Coarse-graining MSMs into macrostates is critical for linking simulation findings to experimental observations.

Purpose of the Study:

• To introduce the minimum variance clustering approach (MVCA) for effective coarse-graining of MSMs.
• To demonstrate MVCA's utility in analyzing protein folding dynamics and comparing simulations across different force fields.

Main Methods:

• Utilized agglomerative clustering with Ward's minimum variance objective function.
• Employed Jensen-Shannon divergence to quantify microstate dynamics similarity.
• Applied MVCA to tripeptide systems, protein folding simulations, and aggregated MSM datasets.

Main Results:

• MVCA produced intuitive results for a tripeptide system and showed robustness against statistical artifacts.
• Identified a misfolded state in one protein folding simulation by revealing different optimal macrostate numbers for distinct force fields.
• Enabled grouping and dynamical similarity quantification of multiple MSMs from various force fields.

Conclusions:

• MVCA provides a robust and intuitive method for coarse-graining MSMs into macrostate models.
• The approach facilitates the identification of distinct conformational states and the comparison of simulation data.
• MVCA is a powerful tool for analyzing dynamical similarity among both model states and dynamical models themselves.