Markov dynamic models for long-timescale protein motion.

Area of Science:

• Computational Biology
• Biophysics
• Protein Dynamics

Background:

• Molecular dynamics (MD) simulations offer atomic-scale insights into protein motion but are computationally demanding and generate vast datasets.
• Analyzing MD data for long-timescale protein dynamics requires efficient modeling techniques.

Purpose of the Study:

• To develop simplified Markov models with hidden states for efficient analysis of long-timescale protein motion from MD simulations.
• To establish a criterion for evaluating model quality based on predictive accuracy for long-timescale protein dynamics.

Main Methods:

• Utilized Markov models with hidden states, where states represent probabilistic distributions over protein conformations.
• Developed a principled evaluation criterion focused on predicting long-timescale protein motions.
• Applied the method to 2D synthetic energy landscapes, alanine dipeptide, and the villin headpiece subdomain (HP-35 NleNle).

Main Results:

• Demonstrated that a simpler three-state Markov model for alanine dipeptide achieved comparable predictive accuracy for long-timescale motions to a widely accepted six-state model.
• Successfully estimated kinetic and dynamic quantities, including mean first-passage time, crucial for protein folding.
• Obtained results consistent with existing experimental measurements.

Conclusions:

• Simplified Markov models offer an efficient and accurate approach to modeling long-timescale protein dynamics from MD data.
• The proposed evaluation criterion effectively assesses model performance for predicting protein motion.
• This methodology provides valuable insights into protein folding kinetics and dynamics.