Discriminant analysis optimizes progress coordinate in weighted ensemble simulations of rare event kinetics

Area of Science:

• Computational biophysics
• Biomolecular simulations
• Machine learning applications

Background:

• Calculating kinetics of rare biomolecular transitions is challenging due to long timescales.
• Standard molecular dynamics simulations are often too slow for these events.
• Weighted ensemble (WE) method enhances sampling but requires careful progress coordinate design.

Purpose of the Study:

• To demonstrate a machine learning approach for designing progress coordinates for WE simulations.
• To improve the efficiency and accuracy of calculating biomolecular conformational transition kinetics.
• To provide a data-driven method with minimal system knowledge requirements.

Main Methods:

• Applied harmonic linear discriminant analysis (HLDA) to build predictive models for class membership.
• Utilized HLDA-derived progress coordinates in WE simulations.
• Tested the method on alanine dipeptide conformational transition and small protein unfolding.

Main Results:

• The machine learning approach successfully designed effective progress coordinates for WE simulations.
• Accurate and efficient computation of conformational transition kinetics was achieved.
• Demonstrated the method's applicability to both small molecules and proteins.

Conclusions:

• Machine learning-guided progress coordinate design enhances WE simulations for biomolecular kinetics.
• This data-driven strategy reduces the need for extensive prior system knowledge.
• The approach shows promise for studying complex, physiologically relevant biomolecular systems.