Enhancing Biomolecular Sampling with Reinforcement Learning: A Tree Search Molecular Dynamics Simulation Method.

Area of Science:

• Computational Chemistry
• Molecular Dynamics Simulations
• Biophysics

Background:

• Molecular simulations are crucial for understanding biomolecular processes.
• Sampling conformational transition pathways is computationally intensive.
• Current methods can get trapped in local energy minima.

Purpose of the Study:

• To develop a novel molecular simulation method to accelerate conformational pathway sampling.
• To improve the efficiency of molecular dynamics simulations.
• To overcome limitations of existing state-of-the-art methods.

Main Methods:

• Introduction of tree search molecular dynamics (TS-MD).
• Application of the Upper Confidence Bounds for Trees (UCT) algorithm, a reinforcement learning technique.
• Utilizing previous simulation results to guide conformational space search.

Main Results:

• TS-MD efficiently searches conformational space.
• The method avoids getting trapped in local stable structures.
• Demonstrated superior performance compared to parallel cascade selection molecular dynamics for miniprotein folding (Chignolin and Trp-cage) in explicit water.

Conclusions:

• TS-MD is an effective method for accelerating the sampling of conformational transition pathways.
• The reinforcement learning approach enhances simulation efficiency.
• TS-MD offers a promising advancement for molecular dynamics studies, particularly in protein folding.