Towards a robust approach to infer causality from molecular dynamics simulations.

Area of Science:

• Computational chemistry
• Molecular dynamics
• Statistical mechanics

Background:

• Distinguishing correlation from causation is crucial in molecular systems.
• Information transfer offers a framework for inferring causal relationships.

Purpose of the Study:

• To probe causality in molecular systems using computational methods.
• To analyze information transfer between collective variables in molecular dynamics simulations.
• To rationalize observed asymmetries in information transfer.

Main Methods:

• Utilizing two independent computational methods based on information transfer.
• Performing molecular dynamics simulations of a single tryptophan in liquid water.
• Analyzing discrete Markov-state and Langevin dynamics on a 2D free energy surface.

Main Results:

• Demonstrated asymmetric information transfer between solute and solvent coordinates in molecular dynamics.
• Observed similar asymmetries in extremely simple systems with equilibrium dynamics.
• Rationalized unidirectional information transfer by free energy landscape and relaxation dynamics asymmetries.

Conclusions:

• Asymmetric information transfer can indicate genuine causal links in molecular systems.
• Model systems help understand information transfer in complex molecular dynamics.
• A computational experiment is proposed to validate causal inference from information transfer.