GPU Computing Delivers Massive Acceleration to Dynamic Protein Simulation

Original Article Date: 2011-01-31 

Using an Electronics Nexus NEWTON GPU Supercomputer, researchers at Purdue University have demonstrated the huge performance benefits that General Purpose Graphical Processing Units (GPGPUs) can bring to scientific modeling.

Nikolai Skrynnikov, Associate Professor of Chemistry, and Yi Xue, Post-Doctoral Research Associate, of Purdue University's Chemistry Department are attempting to understand the dynamic behaviour of disordered proteins, large molecules occuring in the body which are heavily implicated in Alzheimer and Parkinson disease as well as various cancers.

"Proteins are typically thought of as uniquely-shaped, well-structured entities," begins Prof Skynnikov. "Often, however, they turn out to be highly disordered, quickly morphing from one shape to another (if we could observe a disordered protein under a microscope, it would resemble a disturbed earthworm).

The trajectory of the protein Ubiquitin as it morphs from compact to extended form

"Until very recently, it was deemed impossible to model disordered proteins via Molecular Dynamics simulation. The problem is that the number of distinct conformations sampled by disordered proteins is too large - we could not realistically hope to capture even a small fraction of them during the course of the conventional short molecular dynamics simulation. 

The Electronics Nexus NEWTON GPU Supercomputer, shortly prior to shipping to Purdue University.
Specs: 2 x Xeon E5420 2.4GHz Quad Core CPUs, Tyan S7025 mainboard, 6x2GB System RAM, Silverstone ST1500W Power, Lian-Li PC-P80 chassis and 4 x Zotac NVIDIA GTX480 1.5GB 480-core GPUs

"With the advent of GPU computing the situation has changed. Our purchase of a four-GPU workstation from Electronics Nexus has led to a dramatic hundred-fold boost in computational resources. As a direct result of this acquisition, Dr. Yi Xue was able to record multiple micro-second long trajectories of denatured ubiquitin and demonstrate - for the first time - that these trajectories faithfully reproduce the actual dynamics of the molecule (as reflected in his experimental data obtained by means of the Nuclear Magnetic Resonance spectroscopy)."

Graph showing the dramatic improvement in simulation rate of the Amber 11 Molecular Dynamic Simulation Suite when using GPUs over conventional CPUs

The above graph shows that using just a single GTX480 GPU has improved simulation performance by as much as 300x compared to a single conventional CPU core. Dr. Xue is confident that as the bugs in the still very experimental Amber 11 code are ironed-out, speed up with multiple GPUs should allow even faster simulations and greater model complexity.

"Explicit molecular dynamic modeling of such proteins," Prof. Skrynnikov says, "will lead to unique insights into their behavior and function and eventually facilitate the development of new pharmaceutical strategies targeting these disorderly molecules."

 If you've enjoyed success in using our high-performance computers in your line of work and would be interested in having a similar feature article written about it, just drop us a line!