Neural Network and Nearest Neighbor Algorithms for Enhancing Sampling of Molecular Dynamics

Area of Science:

• Computational Chemistry
• Biophysics
• Machine Learning

Background:

• Molecular dynamics (MD) simulations are crucial for studying complex chemical and biological systems.
• Calculating free energy landscapes is often inefficient due to numerous local minima and high energy barriers.
• Enhanced sampling methods like metadynamics improve efficiency but are limited by the number of collective variables (CVs).

Purpose of the Study:

• To develop a novel high-dimensional bias potential method for enhanced sampling in molecular dynamics.
• To overcome the limitations of existing methods regarding the number of collective variables (CVs).
• To enable efficient and ergodic free energy calculations for complex systems.

Main Methods:

• Proposed a high-dimensional bias potential method named NN2B.
• Utilized two machine learning algorithms: nearest neighbor density estimator (NNDE) and artificial neural network (ANN).
• Constructed the bias potential iteratively using short biased MD simulations that account for CV correlations.

Main Results:

• Achieved ergodic sampling in complex chemical and biological systems.
• Successfully calculated free energy landscapes using up to an 8-dimensional bias potential.
• Demonstrated the capability of the NN2B method to handle high-dimensional collective variables.

Conclusions:

• The NN2B method significantly enhances the efficiency of free energy calculations in MD simulations.
• This approach overcomes the dimensionality limitations of traditional enhanced sampling techniques.
• NN2B offers a powerful new tool for exploring complex free energy landscapes in chemistry and biology.