Extensible and Scalable Adaptive Sampling on Supercomputers

Area of Science:

• Computational biology
• Biophysics
• Molecular dynamics

Background:

• Accurate protein dynamics sampling is crucial but challenging with standard molecular dynamics (MD) simulations.
• High-performance computing (HPC) systems are utilized, but "brute force" MD remains time-prohibitive.
• Adaptive sampling methods offer significant speedups (over 10x) compared to standard MD.

Purpose of the Study:

• To present the ExTASY framework for simplifying adaptive sampling on HPC.
• To enable domain experts to efficiently utilize advanced sampling techniques.
• To demonstrate the framework's capability in predicting protein folding dynamics.

Main Methods:

• Developed the ExTASY framework for setting up adaptive sampling strategies.
• Executed adaptive sampling workflows at scale on HPC platforms.
• Applied the framework to predict the folding dynamics of four proteins without prior information.

Main Results:

• ExTASY facilitates the reliable execution of adaptive sampling on HPC.
• The framework significantly enhances the efficiency of protein dynamics simulations.
• Accurate folding dynamics for four proteins were successfully predicted.

Conclusions:

• The ExTASY framework democratizes advanced protein dynamics simulations on HPC.
• Adaptive sampling, enabled by ExTASY, provides a powerful alternative to standard MD.
• This approach accelerates discovery in protein folding and dynamics research.