Understanding the dynamical behavior of large biological systems, e.g., proteins, is a central goal of computational biologists as they seek to understand molecular recognition and catalysis. Locating the minimum free energy path between important states of biomolecular systems yields insight into how often a transition will occur and mechanistic information about how the transition takes place. Traditional molecular dynamics can fail to sample these important transitions in a reasonable computation time as they are thermally limited to low energy states on the complex free energy surface of proteins and other biopolymers. Application of graphics processing units (GPUs) significantly accelerates these computationally intensive calculations, offering a base multiplier for any other enhanced sampling strategy that can be implemented on their advanced architecture.<br/>The Nudged Elastic Band (NEB) approach which applies a series of harmonic restraints between replicas in combination with simulated annealing has proven successful in determining minimum energy pathways for rugged energy surfaces. In NEB, a collection of simulation replicas is first generated along a putative pathway and evolves into a discrete representation of minimum energy transition pathway by stringent energy minimization of the restraints between replicas. We have extended the partial nudged elastic band (PNEB) method to the PMEMD engine to operate with the CUDA version of the program. We designed a shuttle transfer system which minimizes the required bandwidth that must traverse a message passing interface (MPI) between GPUs, and our code also takes advantage of pinned and coalesced memory accesses. Running on NVIDIA GTX-1080TI GPUs in parallel, the Amber18 GPU-accelerated PNEB offers more than 20 times faster sampling than a two-core Intel E5-2620 CPU processor tasked with the same problem. Numerical accuracy is uncompromised. The new implementation is easily extendable to other chain-of-states methods for searching minimum energy pathways such as the string method, and greatly enhances the study of transitional properties of biomolecules in a multidimensional phase space of thousands of atoms–which was otherwise impossible to study with current CPU architectures. Our implementation of NEB for multi-GPU execution within the AMBER software suite could aid computational scientists in developing new drug compounds and novel materials by applying these powerful algorithms on commodity hardware.

Implementation of GPU-accelerated partial nudged elastic band method into Amber for finding minimum free energy pathways in biomolecular systems

256th National Meeting (2018)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.