IKI - MSR Research Workshop
Space Research Institute (IKI), Moscow, Russia
11 June, 2009
Speaker: Yaroslav Kholodov, Department of Applied Mathematics, Moscow Institute of Physics and Technology
Large-scale supramolecular assemblies play fundamental roles in biology. Protein fibers, such as microtubules, actin filaments, fibrin fibers and fibronectin, perform mechanical functions across biological systems, and understanding and control of the response of protein fibers to mechanical tension constitute major areas of research in biology and biophysics. Although molecular simulations are indispensable for the exploration of the energy landscape of biomolecular assemblies, accessing their sub-molecular level behavior under physiological force loads is challenging even for distributed computing systems because of their large size (~1000?10000 of residues). Graphics Processing Units (GPUs) are emerging as a programming platform that provides unprecedented computational power for scientific applications. A modern GPU is a highly parallel processor with high arithmetic and memory bandwidth that enables researchers to carry out experiments in silico at the petascale level. GPU-based calculations require only a modest increase in hardware cost and are 10?100 times faster than heavily optimized CPU-based methods. Recently, NVIDIA introduced CUDA, a general purpose parallel computing environment. A major bottleneck in using GPUs for advanced scientific computation is that CPU-based methods cannot be easily translated to GPUs due to fundamental differences in processor and memory architecture. This problem can be overcome through the development of novel computational methodology for efficient simulations of biomolecular assemblies fully implemented on GPUs. This methodology will be used to explore the unfolding micromechanics of fibrin fibers (FFs) and microtubules (MTs) – structural components of blood clots and the cytoskeleton, respectively. Because all-atom Molecular Dynamics (MD) simulations are currently limited to a 10nm length scale and 100ns duration, whereas the size of FFs and MTs is ~50?400nm and their structural transitions occur on ms?s timescales, all-atom modeling cannot be used to access these transitions. For this reason, coarse-grained (CG) description of proteins and Langevin Dynamics (LD) simulations will be employed. The proposed research will lead to the development of much needed (1) GPU-based computational infrastructure (algorithms, numerical integration schemes, generators of pseudo-random numbers), and (2) CUDA based software for efficient LD simulation of biomolecules, which will be used to provide physical insights into the mechanisms governing the unfolding transitions in FFs and MTs. In addition, GPU-based methodology for LD simulations will enable researchers (3) to explore many other macromolecular assemblies of biological interest (plant and animal viruses, motor proteins, bacterial flagella, etc.) Outcomes from this collaborative project (4) will enhance the predictive power of advanced computer models applied to increasingly larger biomolecular assemblies from systems biology, and (5) will provide conceptual guidance for further development of computer models and theory for describing large-scale molecular assemblies of biological interest, and (6) will lead to new avenues for research.