DeepWEST: Deep Learning of Kinetic Models with the Weighted Ensemble Simulation Toolkit for Enhanced Sampling

Area of Science:

• Computational chemistry
• Biophysics
• Machine learning

Background:

• Molecular dynamics (MD) simulations are advancing but still limited in observing biomolecular conformational transitions.
• Enhanced sampling techniques like the weighted ensemble (WE) method are used to study these transitions.
• The efficiency of WE simulations depends heavily on the initial sampling of the potential energy surface.

Purpose of the Study:

• To introduce deep-learned kinetic modeling to improve the initial state distribution for WE simulations.
• To overcome limitations of traditional MD and WE methods in sampling biomolecular processes.
• To enable more accurate estimation of kinetic rate constants and free energy landscapes.

Main Methods:

• Utilizing deep learning to extract statistically relevant information from short, unbiased MD trajectories.
• Developing a hybrid approach combining deep learning with the weighted ensemble method.
• Identifying metastable states and refining the free energy landscape.

Main Results:

• The hybrid approach provides a well-sampled initial state distribution for WE simulations.
• This method overcomes statistical bias, leading to a more refined free energy landscape.
• The approach efficiently samples kinetic properties, including rate constants.

Conclusions:

• Deep-learned kinetic modeling offers a powerful enhancement for WE simulations.
• This hybrid strategy improves the accuracy and efficiency of studying biomolecular dynamics.
• The method provides a more reliable pathway to understanding complex biological processes.