Effect of Clustering Algorithm on Establishing Markov State Model for Molecular Dynamics Simulations.

Area of Science:

• Computational chemistry
• Biophysics
• Molecular dynamics simulations

Background:

• Markov state models (MSMs) are increasingly used for kinetic analysis of molecular dynamics (MD) simulations.
• Discretization of conformational space is a critical challenge in building accurate MSMs.
• The influence of different clustering strategies on MSM output is not well understood.

Purpose of the Study:

• To investigate how various clustering algorithms affect the construction and output of Markov state models.
• To compare the performance of different clustering strategies in partitioning conformational space for kinetic analysis.
• To identify optimal clustering approaches for characterizing protein dynamics.

Main Methods:

• Application of diverse clustering algorithms (geometric, kinetic, Bayesian, hierarchical) to partition conformational space.
• Construction of Markov state models based on different clustering classifications.
• Calculation of net flux and transition rates between distinct states using the developed MSMs.
• Analysis of protein dynamics for fatty acid binding protein 4 (FABP4) and epidermal growth factor receptor (EGFR).

Main Results:

• Different clustering algorithms yield varied classifications of conformational space.
• Both geometric and kinetic clustering methods demonstrate comparable performance.
• The characteristics of the data significantly influence the construction and results of MSMs.
• A combination of Bayesian and hierarchical clustering proves effective for identifying metastable states.

Conclusions:

• The choice of clustering strategy critically impacts Markov state model outcomes in MD simulations.
• Geometric and kinetic clustering are viable options, performing similarly.
• Bayesian and hierarchical clustering offer a robust approach for identifying metastable states in complex biomolecular systems.