Folding a small protein using harmonic linear discriminant analysis.

Area of Science:

• Computational Chemistry
• Biophysics
• Molecular Dynamics

Background:

• Many scientific processes involve long time scales, limiting standard simulation techniques.
• Enhanced sampling methods, particularly those using collective variables, are crucial for studying these processes.
• Designing effective collective variables is a significant challenge in applying these methods.

Purpose of the Study:

• To introduce and evaluate Harmonic Linear Discriminant Analysis (HLDA) for systematic collective variable construction.
• To assess the efficiency of HLDA in capturing the essential physics of molecular processes.
• To apply HLDA to the complex problem of chignolin folding in water.

Main Methods:

• Developed Harmonic Linear Discriminant Analysis (HLDA) to construct collective variables from descriptors.
• Utilized short, unbiased molecular dynamics simulations to gather input data for HLDA.
• Applied the constructed collective variable in metadynamics simulations, combined with parallel tempering for convergence.
• Investigated the impact of different descriptor sets on collective variable performance.

Main Results:

• The one-dimensional collective variable derived from HLDA revealed key physics of chignolin folding prior to biased simulations.
• HLDA-based metadynamics simulations successfully captured the folding and unfolding of chignolin.
• Combined metadynamics and parallel tempering yielded converged simulation results.
• The performance of the collective variable was examined with varying descriptor sets.

Conclusions:

• HLDA provides a systematic and efficient approach to constructing collective variables for enhanced sampling.
• The method successfully captures the underlying physics of complex molecular processes like protein folding.
• HLDA is a valuable tool for advancing molecular simulations of long-timescale phenomena.