GPUs have significantly more computational cores than central processing units (CPUs). Molecular dynamics simulations, which rely on force calculations that are repetitive and parallelizable, have taken advantage of GPUs to simulate large complex systems such as protein interactions in explicit macromolecular crowding environments, and improved the modeling of dynamic properties such as solvent accessible surface area. With an NIH supplement we will purchase a dedicated GPU cluster to advance the work of our parent grant in elucidating the sources and connections between altered translation-elongation kinetics of protein synthesis and its impact on protein structure and function. Our primary computational tool involves simulating protein synthesis by the ribosome, and a typical trajectory takes 60 days to run on CPUs. We have found we can speed up these simulations 20 to 30-fold acceleration in computation time for protein synthesis utilizing GPUs over CPUs and thus what normally would have taken two months can be done in as little as two days. It also opens up the ability to run more trajectories and thereby obtain better statistical power (i.e., smaller error bars). By having our own GPU cluster the research aims of the parent grant will be greatly accelerated.
Advances in molecular dynamics software and the use of GPUs has increased the speed with which complex biological systems can be modeled. We have found that we can run our simulations 20 to 30- fold faster using GPUs. Therefore, a small dedicated GPU cluster will vastly increase our research capabilities under the parent grant.
| Samelson, Avi J; Bolin, Eric; Costello, Shawn M et al. (2018) Kinetic and structural comparison of a protein's cotranslational folding and refolding pathways. Sci Adv 4:eaas9098 |
