High-throughput all-atom molecular dynamics simulations using distributed computing.

Area of Science:

• Computational chemistry
• Biophysics
• Molecular modeling

Background:

• Molecular dynamics simulations are crucial for modeling macromolecular systems but are computationally expensive, requiring high-performance computing (HPC) resources.
• Innovations in graphics processing unit (GPU) acceleration offer a potential solution to reduce computational costs.

Purpose of the Study:

• To review innovations in accelerating molecular dynamics simulations on GPUs.
• To describe the GPUGRID volunteer computing project.
• To demonstrate GPUGRID's capability for high-throughput, accurate binding affinity prediction.

Main Methods:

• Utilizing the GPU resources of non-dedicated desktop and workstation computers through the GPUGRID project.
• Simulating thousands of all-atom molecular trajectories at an average of 20 ns/day for systems of 30,000-80,000 atoms.
• Employing a potential of mean force (PMF) protocol with umbrella sampling to compute binding free energies.

Main Results:

• Demonstrated the simulation of numerous molecular trajectories, achieving high throughput.
• Successfully computed accurate binding affinities for the Src SH2 domain/pYEEI ligand complex.
• Obtained a standard free energy of binding of -8.7 ± 0.4 kcal/mol, closely matching experimental results (within 0.7 kcal/mol).

Conclusions:

• GPUGRID provides a robust infrastructure for accelerating molecular dynamics simulations using volunteer computing.
• This approach enables high-throughput and accurate prediction of binding affinities.
• The developed system has significant implications for drug discovery and molecular modeling research.