Exploring Protocols to Build Reservoirs to Accelerate Temperature Replica Exchange MD Simulations

Area of Science:

• Computational chemistry
• Biophysics
• Molecular dynamics simulations

Background:

• Temperature replica exchange molecular dynamics (REMD) is a key technique for accelerating biomolecular simulations.
• Existing REMD variants, like reservoir REMD (RREMD), offer improved conformational sampling but face challenges in reservoir construction and GPU implementation.
• These limitations have hindered the widespread adoption of advanced REMD methods.

Purpose of the Study:

• To optimize reservoir REMD (RREMD) for enhanced sampling in molecular dynamics.
• To address the computational bottlenecks of RREMD by porting it to GPUs.
• To evaluate different reservoir construction protocols and their impact on simulation accuracy and efficiency.

Main Methods:

• Porting the Amber RREMD code to run on Graphics Processing Units (GPUs), achieving a 20x speedup over CPU.
• Developing and testing protocols for constructing both Boltzmann-weighted and non-Boltzmann reservoirs.
• Comparing RREMD simulations with conventional temperature-based REMD for accuracy and convergence rates.

Main Results:

• The GPU-accelerated RREMD achieved a 20x speed increase compared to the CPU version.
• RREMD simulations using the recommended protocols accurately reproduced Boltzmann-weighted ensembles.
• RREMD demonstrated at least 15x faster convergence rates than conventional REMD, even for proteins larger than 50 amino acids.

Conclusions:

• GPU-accelerated RREMD provides a significant enhancement in conformational sampling efficiency for biomolecular simulations.
• The developed protocols enable accurate and accelerated simulations, overcoming previous limitations of RREMD.
• This optimized RREMD method offers a powerful tool for studying large proteins and achieving faster convergence.