Generative models of conformational dynamics.

Area of Science:

• Computational Biology
• Biophysics
• Statistical Mechanics

Background:

• Atomistic simulations, including Molecular Dynamics and Monte Carlo, are crucial for studying protein conformational dynamics.
• Analysis of simulation data reveals thermodynamic and kinetic properties of biological systems.
• Generative models capture the joint probability distribution of atomic behaviors in molecular systems.

Purpose of the Study:

• To review traditional and emerging methods for learning generative models from atomistic simulation data.
• To explain how generative models elucidate atomic correlations and predict responses to structural changes.
• To introduce GAMELAN as a novel approach for complex conformational dynamics.

Main Methods:

• Review of traditional methods yielding multivariate Gaussian models.
• Introduction of GAMELAN (GRAPHICAL MODELS OF ENERGY LANDSCAPES).
• Application of methods to long timescale simulation data.

Main Results:

• Traditional methods generate Gaussian models of protein dynamics.
• GAMELAN enables the creation of generative models for complex, non-Gaussian dynamics.
• These models reveal intricate correlation structures between atoms.

Conclusions:

• Generative models are powerful tools for understanding protein conformational dynamics.
• GAMELAN offers a significant advancement for modeling complex biological processes like allostery, binding, and folding.
• The developed methods enhance predictions of system behavior under perturbation.