Quantitative comparison of adaptive sampling methods for protein dynamics.

Area of Science:

• Computational chemistry
• Biophysics
• Molecular dynamics simulations

Background:

• Adaptive sampling methods accelerate rare event simulations in molecular dynamics (MD).
• Markov state models are often combined with adaptive sampling.
• Optimal strategies for diverse physical systems remain unclear.

Purpose of the Study:

• Systematically evaluate adaptive sampling strategies for protein folding simulations.
• Compare strategy performance with and without prior system knowledge.
• Determine theoretical limits for sampling speed-up.

Main Methods:

• Emulated adaptive sampling strategies using models from existing MD trajectories.
• Evaluated strategies on fast-folding proteins.
• Assessed performance based on simulation goals (e.g., sampling slow dynamics vs. exploring conformational space).

Main Results:

• No single adaptive sampling strategy is optimal for all goals.
• Metastable region identification is most efficient for sampling slow dynamics without prior knowledge.
• Microstate identification is superior for exploring new conformational regions.
• Maximum speed-up increases with protein folding time.

Conclusions:

• The choice of adaptive sampling strategy depends on the specific simulation objective.
• Metastable region-based strategies are effective for unbiased sampling of slow processes.
• Microstate-based strategies enhance exploration of conformational landscapes.
• Adaptive sampling shows promise for characterizing slower biological processes beyond fast-folding proteins.