Automated bias potential optimization via position and variance control in umbrella sampling.

Area of Science:

• Computational Chemistry
• Molecular Dynamics Simulations
• Statistical Mechanics

Background:

• Free-energy landscapes (FELs) are vital for understanding molecular processes.
• Standard umbrella sampling (US) has limitations in sampling high-energy regions and controlling window distributions.
• Previous work introduced adaptive bias potential optimization for FEL sampling.

Purpose of the Study:

• To refine an optimization-based approach for enhanced free-energy landscape sampling.
• To explicitly control both positions and variances of sampling distributions in umbrella sampling.
• To improve the accuracy and convergence of FEL reconstruction.

Main Methods:

• Developed an optimization method controlling sampling distribution positions and variances using target Gaussian distributions with variance upper bounds.
• Applied the method to Langevin dynamics simulations of the Wolfe-Quapp potential.
• Validated the approach using molecular dynamics simulations of alanine dipeptide in water.

Main Results:

• The optimized method demonstrated superior convergence compared to non-optimized simulations.
• Achieved significantly improved accuracy and faster convergence in reconstructing FELs, particularly near saddle points and steep gradients.
• The optimized simulations outperformed standard US in accuracy and speed.

Conclusions:

• The refined optimization framework offers a robust and generalizable strategy for automated bias potential tuning.
• Facilitates accurate FEL reconstruction in complex molecular systems, including biomolecules.
• The method is integrated with the PLUMED package and available on GitHub for broad accessibility.