On inference of causality for discrete state models in a multiscale context

Area of Science:

• Computational Chemistry
• Statistical Mechanics
• Data Science

Background:

• Discrete state models, including Markov state models, are crucial for analyzing molecular dynamics (MD).
• Existing models face challenges with systematically missing data, leading to nonstationary and nonhomogeneous inference problems.
• There is a need for methods to identify optimal and simple discrete state models under such complex conditions.

Purpose of the Study:

• To extend discrete state models to handle systematically missing scales in data.
• To develop a framework for identifying simultaneously optimal and simple discrete state models.
• To apply these methods to molecular dynamics data for understanding causality.

Main Methods:

• Utilizing nonstationary data analysis and information theory tools.
• Developing a formalism for inferring discrete state models from incomplete data.
• Applying the framework to coarse-grained molecular dynamics torsion angle data of polypeptides.

Main Results:

• Successfully identified optimal and simple discrete state models for nonstationary MD data.
• Demonstrated that the optimal model for polypeptide torsion angles exhibits localized spatial and temporal causality.
• The findings suggest new approaches for MD analysis using percolation theory and subgridscale modeling.

Conclusions:

• The developed methods effectively address missing data challenges in discrete state modeling for MD.
• Understanding localized causality in MD is crucial for accurate process interpretation.
• This work opens avenues for integrating advanced modeling techniques into molecular dynamics simulations.