Using Time Dependent Rate Analysis to Evaluate the Quality of Machine Learned Reaction Coordinates for Biasing and...

Area of Science:

• Computational Chemistry
• Molecular Dynamics
• Biophysics

Background:

• Accurate reaction coordinates (RCs) are crucial for molecular kinetics.
• Quantitative metrics for evaluating RC quality are limited.
• The Exponential Average Time-dependent Rate (EATR) method offers a potential solution.

Purpose of the Study:

• To evaluate the dimensionless gamma (γ) metric for assessing RC quality.
• To determine if γ can guide the iterative improvement of RCs.
• To establish γ as a practical criterion for kinetic prediction.

Main Methods:

• Utilized the γ metric from the EATR method.
• Employed the State Predictive Information Bottleneck (SPIB) approach for RC approximation.
• Evaluated RCs across six protein-ligand dissociation systems.
• Computed γ values and mean accelerated times (τ̅_accel).

Main Results:

• Demonstrated that γ increases as RCs are iteratively updated with new data.
• Observed a consistent inverse correlation between γ and τ̅_accel.
• Showed that higher γ values correspond to shorter accelerated times.

Conclusions:

• The γ metric is a practical and effective criterion for evaluating RCs.
• γ provides guidance for selecting optimal SPIB-derived coordinates.
• This facilitates more accurate quantitative kinetic predictions in molecular simulations.