Enhanced Sampling of Biomolecular Slow Conformational Transitions Using Adaptive Sampling and Machine Learning

Area of Science:

• Computational chemistry and biophysics
• Molecular dynamics simulations
• Enhanced sampling techniques

Background:

• Biomolecular simulations face the "time scale problem" for rare events.
• Enhanced sampling methods often require predefined collective variables (CVs).
• Machine learning approaches typically need extensive data covering the entire free energy surface (FES).

Purpose of the Study:

• To develop an automated iterative pipeline to address limitations in enhanced sampling.
• To enable efficient exploration of rare events and FES in biomolecular simulations.
• To reduce the reliance on predefined collective variables and extensive training data.

Main Methods:

• Utilizes a CV-free count-based adaptive sampling for rare event data generation.
• Employs Koopman-reweighted time-lagged independent component analysis (KTICA) to identify slow modes.
• Leverages on-the-fly probability enhanced sampling (OPES) for efficient FES exploration.

Main Results:

• Successfully generated data rich in rare events without CVs.
• Identified slow dynamic modes using KTICA.
• Demonstrated efficient FES exploration with OPES.
• Validated the pipeline on alanine dipeptide (Ala2) and deca-alanine (Ala10) systems.

Conclusions:

• The developed automated pipeline effectively mitigates the time scale problem in biomolecular simulations.
• The method enhances the exploration of rare events and FES, reducing the need for CVs.
• The pipeline shows broad applicability across different biomolecular systems, offering an advantage over traditional Markov State Model (MSM) approaches.