Enhanced sampling in explicit solvent by deep learning module in FSATOOL.

Area of Science:

• Computational chemistry
• Biomolecular simulations
• Machine learning

Background:

• FSATOOL previously lacked explicit solvent simulation capabilities.
• Existing molecular dynamics engine was limited to vacuum or implicit solvent simulations.

Purpose of the Study:

• To enhance FSATOOL by implementing explicit solvent simulations.
• To integrate deep learning for advanced biomolecular trajectory analysis.
• To improve enhanced sampling methods for conformational transitions.

Main Methods:

• Implementation of the smooth particle mesh Ewald (SPME) method for explicit solvent simulations.
• Development of a deep learning module within FSATOOL.
• Integration of state-free reversible VAMPnets and time-lagged autoencoders for collective variable identification.
• Utilizing identified collective variables for bias potential construction in enhanced sampling.

Main Results:

• A new molecular dynamics engine supporting explicit solvent simulations on both CPU and GPU.
• Successful integration of deep learning-based analysis tools.
• Demonstrated utility of VAMPnets and time-lagged autoencoders in identifying key collective variables for biomolecular transitions.
• FSATOOL's enhanced sampling capabilities were validated through simulations.

Conclusions:

• FSATOOL is now a more versatile tool for molecular simulations, including explicit solvent dynamics.
• The new deep learning module significantly advances the analysis of biomolecular conformational changes.
• The implemented methods provide powerful approaches for enhanced sampling and understanding molecular mechanisms.