Probabilistic Forecasting for Coarse-Grained Molecular Dynamics.

Area of Science:

• Computational Chemistry
• Biophysics
• Machine Learning

Background:

• Coarse-grained molecular dynamics (CGMD) offers computational efficiency for large systems.
• However, CGMD often struggles with atomistic kinetics, especially with memory effects and slow dynamics.
• Existing methods may not fully capture non-Markovian dynamics crucial for accurate biomolecular simulations.

Purpose of the Study:

• To introduce a novel machine learning framework, Probabilistic Forecasting for Coarse-Graining (PFCG).
• To enable learning stochastic coarse-grained equations of motion directly from atomistic data.
• To improve the accuracy of coarse-grained simulations by incorporating non-Markovian effects.

Main Methods:

• Developed PFCG, a machine learning framework for coarse-grained simulations.
• Formulated coarse-grained simulation as a probabilistic time-series forecasting problem.
• Incorporated non-Markovian effects using finite trajectory history without explicit memory kernels.

Main Results:

• PFCG successfully applied to miniproteins and polyalanine peptides.
• Non-Markovian PFCG models significantly improved dynamical agreement with atomistic simulations compared to Markovian baselines.
• Models maintained excellent agreement with stationary distributions and showed robustness under sparse sampling.

Conclusions:

• PFCG effectively learns stochastic coarse-grained equations of motion from atomistic trajectories.
• The framework enhances dynamical fidelity in coarse-grained simulations by capturing non-Markovian contributions.
• PFCG represents a valuable complementary approach to existing machine learning-based coarse-graining methods for biomolecular modeling.