Recent advances in machine learning and coarse-grained potentials for biomolecular simulations.

Area of Science:

• Computational Biology
• Biophysics
• Molecular Dynamics

Background:

• Biomolecular simulations are vital for understanding biological systems, from drug discovery to virus-host interactions.
• All-atom (AA) simulations offer high resolution but are computationally limited to short timescales.
• Coarse-grained (CG) models extend simulation scales but often lack atomic accuracy.

Purpose of the Study:

• To review recent advancements in machine learning (ML)-driven biomolecular simulations.
• To highlight the integration of ML with CG models for enhanced accuracy and extended scales.
• To address the persistent challenge of parameterizing reliable CG potentials.

Main Methods:

• Development of ML potentials with quantum-mechanical accuracy.
• ML-assisted backmapping techniques from CG to AA resolutions.
• Discussion of widely used CG potentials and their integration with ML.

Main Results:

• ML potentials achieve quantum-mechanical accuracy, improving simulation reliability.
• ML-assisted backmapping bridges the gap between CG and AA resolutions.
• Integration of ML and CG approaches significantly enhances simulation accuracy and extends accessible time and length scales.

Conclusions:

• ML-driven approaches are overcoming key limitations in biomolecular simulations.
• The synergy between ML and CG models offers a powerful strategy for studying complex biological systems.
• These advancements promise to accelerate discoveries in areas like drug development and molecular virology.