Transferable generative models bridge femtosecond to nanosecond time-step molecular dynamics

Area of Science:

• Computational Chemistry
• Molecular Dynamics
• Biophysics

Background:

• Molecular dynamics simulations are crucial for understanding molecular structure, dynamics, and reactivity.
• Conventional simulations are limited by femtosecond time steps, restricting access to slow processes governing chemical function.

Purpose of the Study:

• To develop a deep generative modeling framework to accelerate molecular dynamics sampling.
• To enable quantitative characterization of equilibrium ensembles and dynamical relaxation processes.

Main Methods:

• Implemented a deep generative modeling framework.
• Applied the framework to small organic molecules and peptides.
• Validated the approach for quantitative characterization and extended timescale dynamics.

Main Results:

• Achieved a four-orders-of-magnitude acceleration in molecular dynamics sampling while maintaining physical realism.
• Enabled quantitative characterization of equilibrium ensembles and dynamical relaxation processes.
• Demonstrated generalization across chemical composition and system size, including extrapolation to larger peptides.

Conclusions:

• The developed framework significantly expands the accessible range of molecular motions without sacrificing atomistic detail.
• Opens new opportunities for probing conformational landscapes, thermodynamics, and kinetics in chemistry and biophysics.
• Provides a powerful tool for studying complex molecular systems on extended timescales.