Machine-Learning Based Stacked Ensemble Model for Accurate Analysis of Molecular Dynamics Simulations.

Area of Science:

• Computational chemistry
• Materials science
• Machine learning

Background:

• Accurate analysis of molecular dynamics (MD) simulations is critical for material discovery and force-field development.
• Existing algorithmic approaches have limitations in handling the automated nature of these processes.

Purpose of the Study:

• To develop and validate a novel computational framework integrating machine learning (ML) with MD simulations.
• To overcome the drawbacks of traditional algorithm-based analysis methods for MD trajectories.

Main Methods:

• Developed and utilized stacked ensemble models (SEMs) for the first time with MD simulations.
• Two SEMs were constructed using layered networks of Random Forest, Artificial Neural Network, Support Vector Regression, Kernel Ridge Regression, and k-Nearest Neighbors models.
• Applied SEMs to analyze the contact angle and hydrogen bonds of a water droplet simulation.

Main Results:

• The SEMs demonstrated higher accuracy compared to individual ML models, as indicated by root-mean-square error values.
• Uncertainty quantification and sensitivity analysis confirmed the robustness and accuracy of the SEMs.
• The framework effectively captured critical information from MD simulation trajectories.

Conclusions:

• The developed computational framework offers a general, robust, and accurate method for analyzing large MD simulation trajectories.
• This approach has significant potential for accelerating material discovery and force-field parameterization.
• The integration of SEMs with MD simulations represents a significant advancement in computational analysis.