Bacterial Protein Expression Core
Cell Biology and Imaging Core
EM Crystallography Core
EM Tomography Core
Eukaryotic Protein Expression Core
Fluorescence Spectroscopy Core
Protein Interactions Core
Protein NMR Spectroscopy Core
RNA Structure and Dynamics Core
Tissue EM Core
Virus Imaging Core
X-ray Crystallography Core
Director: Greg Voth, PhD
The Computation Core provides leading edge theory and simulation capability in atomistic molecular dynamics, coarse-graining, and multiscale modeling. Core expertise is available to all CHEETAH participants, and our current focus is on computational analyses of the assembly and stability of immature virions and capsids. Toward this end, we will carry out multiscale computer simulations of protein-protein interactions and dynamics and develop new coarse-graining and multiscale modeling methods. The Computation Core is housed in newly renovated space in the Searle Chemistry Laboratory and the Gordon Center for Integrative Science on the University of Chicago campus.
Instrumentation and Computing Resources
The Computation Core has a 768-core Dell PE R610 rackmount server cluster. It has 96 compute nodes, a login/management node, a fully non-blocking Qlogic DDR InfiniBand (IB) switch, a Gigabit Ethernet (GigE) switch, and an 8-TB global scratch space. Each compute node has two 2.66 GHz quad-core Intel Xeon processors (for a total of 8 cores per node), 24 GB RAM (3 GB per core), 80/160 GB local disk, and an x4 DDR IB Qlogic dual-port HCA (Host Channel Adapter) card over x8 PCI-E for very low-latency and high-bandwidth communication. A 96-core, 16 GPU Graphical Processing Unit (GPU) cluster has also recently been purchased. This cluster currently consists of 8 compute nodes, a login/management node, 16 GPU computational cards (nVidia Tesla M2050 and M2070 -- 2 assigned per node), a fully non-blocking Mellanox QDR IB switch, and a GigE switch. Each compute node has two 2.66 GHz hex-core Intel Xeon processors (for a total of 12 cores per node), 24 GB RAM, 250GB local disk, an x4 QDR IB Mellanox dual-port HCA, and an Nvidia Host Interface Card (HIC) over x16 PCI-E for connectivity to the GPU computational cards. During the current funding cycle, we will add an additional 4 compute nodes in a single Dell PowerEdge 6100, along with 8 NVIDIA Tesla M2070 PCIe x16 GPGPU Cards (housed in a Dell PowerEdge C410x chassis).
National Supercomputing Resources
The Computation Core also has substantial yearly allocations on NSF and DOE supercomputer systems (33 million node hours in total in the present year for NSF XSEDE computers, 130 million node hours for a multi-year DOE Early Science Project for the Argonne 10-petaFLOPS 768,000 processor IBM Blue Gene/Q computer scheduled to be delivered in 2012), some of which are to be devoted to the simulations described in this project (and see the letters of support from Fahey and Stevens). Voth is also PI (with B. Roux as co-PI) on an NSF PRAC Award for the 7.62-petaFLOPS 388,608 processor Blue Waters Cray computer at University of Illinois NCSA scheduled to be delivered in 2012, with an initial allocation of 90 million node hours.
Many problems in biology (including those related to the CHEETAH Specific Aims) involve phenomena that are intrinsically multiscale. Such phenomena may span six to nine orders of magnitude in length and time scales. The Computation Core has developed and applied multiscale simulation and modeling concepts that tie together atomistic and coarse-grained (CG) simulation, especially for membranes, membrane-bound proteins, and multi-protein assemblies such as the HIV-1 capsid and immature virion. These methods and their application will be expanded in the next project period. The Computation Core also has extensive capabilities for atomistic molecular dynamics (MD) simulation, including expertise with all of the main biomolecular MD codes such as CHARMM, AMBER, NAMD, GROMACS, and LAMMPS, as well as ab initio quantum chemistry and QM/MM codes such as GAUSSIAN and CP2K for use in force field development. The Core has further developed the multiscale coarse-graining (MS-CG) method that systematically formulates CG models from atomistic-scale MD force data.