Spectral Map for Slow Collective Variables, Markovian Dynamics, and Transition State Ensembles

Area of Science:

• Physical chemistry
• Computational chemistry
• Statistical mechanics

Background:

• Complex molecular systems exhibit long-time dynamics crucial for their characteristics.
• Simplifying dynamics involves identifying slow collective variables (CVs) and treating fast variables as noise.
• This reduction allows modeling dynamics as diffusion on a free-energy landscape, approximating Markovian behavior.

Purpose of the Study:

• Advance the spectral map technique for learning slow CVs.
• Apply the enhanced framework to a high-dimensional protein folding process.
• Demonstrate the utility of learned CVs for understanding molecular dynamics and protein folding.

Main Methods:

• Utilized spectral map, a statistical learning technique maximizing spectral gaps of transition matrices.
• Implemented Markov transition matrix coarse-graining to partition the CV space kinetically.
• Defined transition state ensembles based on the partitioned CV space.

Main Results:

• Learned slow CVs closely approximate the Markovian limit for overdamped diffusion.
• Coordinate-dependent diffusion coefficients showed minimal impact on free-energy landscapes.
• Spectral maps effectively quantified feature importance and compared learned CVs with traditional structural descriptors.

Conclusions:

• A single slow CV identified by spectral map can serve as a physical reaction coordinate.
• The technique accurately captures essential characteristics of protein folding dynamics.
• Spectral map offers a powerful tool for analyzing complex molecular systems.