Identification of kinetic order parameters for non-equilibrium dynamics

Fabian Paul, Hao Wu, Maximilian Vossel, Bert L. de Groot, Frank NoéThe Journal of Chemical Physics 2019 · DOI

TL;DR

This paper introduces the Variational Approach for Markov Processes (VAMP) as a powerful new tool for understanding complex molecular dynamics, especially when systems are not in equilibrium. VAMP overcomes limitations of existing methods like TICA by accurately identifying key "order parameters" that govern slow, important processes, even in systems that don't obey detailed balance.

What Problem Does This Paper Address?

Analyzing the behavior of complex systems, like large molecules or materials, often involves simplifying their high-dimensional dynamics into a few key "collective variables" or "order parameters." These parameters help us understand slow processes, identify distinct states, and build simplified models like Markov State Models (MSMs). Traditional methods, such as Time-lagged Independent Component Analysis (TICA), are excellent for this but rely on a crucial assumption: the system must be in equilibrium and obey "detailed balance" (meaning processes are reversible). Many real-world systems, however, are inherently out of equilibrium (e.g., driven by external forces like electric fields) or are studied with short, non-equilibrium simulations. Applying TICA to such data leads to significant errors. This paper addresses the need for a dimension reduction method that works reliably for non-equilibrium dynamics and data.

Methods

The core of this work is the application and validation of the Variational Approach for Markov Processes (VAMP). VAMP reformulates the problem of finding order parameters as a regression problem, dispensing with the assumptions of stationarity and detailed balance. It leverages the Koopman operator framework but uses a restricted singular value decomposition (SVD) of a regression matrix derived from simulation data. This allows VAMP to identify a low-dimensional space of order parameters that minimize a regression error. The authors also extend the Chapman-Kolmogorov test to validate the Markov property of the dimensionality-reduced models obtained with VAMP, and they use PCCA+ (Probabilistic Clustering of Coherent Structures) for spectral clustering to identify long-lived states from the VAMP components.

Datasets

The authors evaluate VAMP on two main systems:

  1. Asymmetric Simple Exclusion Process (ASEP): A well-established toy model for single-file diffusion, used to generate simulated trajectories representing non-equilibrium dynamics.
  2. KcsA Potassium Channel Molecular Dynamics (MD) Simulations: Real-world MD data of a potassium channel protein subjected to an external electrochemical potential, creating a non-equilibrium system. This dataset consists of 15.1 µs of simulation data.

Key Insights and Ideas

  • VAMP is a direct replacement for TICA in non-equilibrium scenarios: The paper strongly advocates for VAMP as a superior alternative to TICA when detailed balance is violated or data is not from equilibrium.
  • Kinetic distance can be reliably computed with VAMP for non-reversible systems: Unlike TICA, VAMP can accurately capture the kinetic distance in systems where processes are not reversible, a crucial aspect for understanding complex dynamics.
  • VAMP components enable robust identification of metastable states: The singular functions identified by VAMP serve as excellent collective variables for clustering and identifying long-lived, distinct states within the system's dynamics, even in non-equilibrium conditions.
  • The Chapman-Kolmogorov test is a valuable tool for validating VAMP-derived models: The authors demonstrate how to adapt this standard test to assess the quality of dimensionality reduction achieved by VAMP.

Key Results

  • ASEP Model: VAMP successfully reproduced time-lagged covariances and auto-covariances for the ASEP, demonstrating its ability to capture dynamics in a non-reversible system. The kinetic distance computed by VAMP was shown to be more accurate than TICA's for this non-reversible model.
  • KcsA Channel: VAMP identified 15 long-lived metastable states in the KcsA channel's conformational dynamics. These states were linked to distinct ion conductances, revealing a direct connection between protein conformation and channel function. The analysis provided insights into the coupling of conformational changes, the selectivity filter configuration, and ion conductance.

Limitations and Future Work

  • Basis Set Dependence: The accuracy of VAMP's dimensionality reduction can depend on the chosen set of input features (basis functions). While VAMP can handle non-linear combinations of features (e.g., via VAMPnets), the linear approach used in this paper relies on a carefully selected set of features.
  • Sampling Limitations: For very rare events or extremely long timescales, even VAMP might require extensive sampling. The authors note that for the KcsA channel, some observed transitions between macro-states were rare, suggesting that longer simulations might reveal more about reversibility.
  • Future Work: The authors suggest exploring more expressive basis sets for VAMP and applying it to a wider range of non-equilibrium systems. They also highlight the potential of VAMP for feature selection and scoring in molecular kinetics. The development of VAMPnets for non-linear order parameters is also an ongoing area.

Summary updated

Abstract

A popular approach to analyze the dynamics of high-dimensional many-body systems, such as macromolecules, is to project the trajectories onto a space of slowly varying collective variables, where subsequent analyses are made, such as clustering or estimation of free energy profiles or Markov state models. However, existing "dynamical" dimension reduction methods, such as the time-lagged independent component analysis (TICA), are only valid if the dynamics obeys detailed balance (microscopic reversibility) and typically require long, equilibrated simulation trajectories. Here, we develop a dimension reduction method for non-equilibrium dynamics based on the recently developed Variational Approach for Markov Processes (VAMP) by Wu and Noé. VAMP is illustrated by obtaining a low-dimensional description of a single file ion diffusion model and by identifying long-lived states from molecular dynamics simulations of the KcsA channel protein in an external electrochemical potential. This analysis provides detailed insights into the coupling of conformational dynamics, the configuration of the selectivity filter, and the conductance of the channel. We recommend VAMP as a replacement for the less general TICA method.
BibTeX
@article{paul2019identification,
  author = {Fabian Paul and Hao Wu and Maximilian Vossel and Bert L. de Groot and Frank Noé},
  title = {{Identification of kinetic order parameters for non-equilibrium dynamics}},
  journal = {The Journal of Chemical Physics},
  year = {2019},
  doi = {10.1063/1.5083627},
}
Figure