Biology

Sequence Analysis, Microarray Data Analysis, Drug Screening and Biomedical Text Mining. (P.I.: Dr. Craig Struble)

Research is pursued in the areas of genome-wide SNP analysis of the human genome for disease associations, phylogeny analysis of RNA sequences extracted from anaerobic digester seeking ways to increase methane production as an alternative energy source and improve digester stability, creation and analysis of networks/graphs from micro array data and screening of library of over 10000 small molecules available for chemical analysis at Marquette. With the help of Accelerators, the time for screening these small molecules is reduced to a large extent and it allows for higher resolution such screens.

Large-Scale Statistical-based Phylogenetics Interface. (P.I.: Dr. Xizhou Feng)

Statistical-based phylogenetic interface tools are essential to evolutionary biology, genomic analysis and biomedical sciences. The PBPI (Parallel Bayesian Phylogenetic Interface) software is a high performance toolkit developed to address the computational challenges in large-scale statistical phylogenetic interface. PBPI uses Markov chain Monte Carlo (MCMC). Accelerators are beneficial in the speedup of PBPI as calculating the likelihood of a chain of tree samples across the entire length of th genomic sequences dominates the computation cost of PBPI and a fine-grained sequence-level parallelism speeds up the calculation. More information at: http://ndssl.vbi.vt.edu/people/fengx.php

Large-Scale Phylogenetic Computations using Genes and Genomes. (P.I.: Dr. Khidir Hilu)

GenBanks have become an invaluable source of information that can be mined in many areas including phylogenetic reconstructions and gene and genome evolution. One of the projects undertaken is the Tree of Life Project. It focuses on land plant diversity and assess the type of genes that can provide the best structure for land plant phylogeny. The genomes studied contain as much as 60000 nucleotide characters. Accelerators help in the reduction of time to re-sample analyses of various species. More information at: http://www.biology.vt.edu/faculty/hilu/Hilu_Lab_Website/Research.html

Evolutionary Genomics and Bioinformatics Tool Development. (P.I.: Dr. Liqing Zhang)

Recent discovery of genes generated by retro position - retro genes has attracted much attention in the research community to understand the evolutionary dynamics of these genes. The process of annotation is computationally very demanding. With the help of Accelerators, we plan to run GENEWISE to map the exon-in-tron structure of a piece of DNAs given in a protein sequence using Dynamic Programming and speed up the annotation process. More information at: http://people.cs.vt.edu/~lqzhang/

Large Scale Neural Modeling. (P.I.: Dr. Scott Beardsley)

The objective of the project is to develop large scale models of neural processing to characterize the representation and extraction of sensorimotor information in the human brain. Testing and evaluating the effectiveness of the system requires modeling of thousands of neurons at millisecond temporal resolution. Access to Accelerators reduces the time to execute these simulations significantly. This would in turn help in the modeling of neuronal processing in the brain and enable the development of applications to visualize neural dynamics across multiple spatial and temporal scales.

Chemistry

Reduced-Scaling Quantum Chemical Methods (P.I.: Dr. T. Daniel Crawford)

Advanced quantum mechanical models such as coupled cluster theory have demonstrated the power of theory and computation by providing hyper-accurate predictions of a vast array of molecular properties. Thus, it is now possible to perform "computational experiments" to simulate chemical events for systems that either cannot be isolated in the lab or are too dangerous. Unfortunately, such models involve steep polynomial scaling (O(N⁷)). With the help of Accelerators, we intend to speed such computations by orders of magnitude More information at: http://www.files.chem.vt.edu/chem-dept/crawford/

Computation of Bimolecular Electrostatic Potential. (P.I.: Dr. Alexey Onufriev)

In its original form, determining the surface potential requires an (O(N2)) all-atom computation, where N is the number of atoms. We calculate electrostatic surface potential using the analytic, linearized Poisson-Boltzmann (ALPB) model. We successfully mapped GEM, an open source software package for computing bimolecular electrostatic potential using the aforementioned ALPB model onto a GPGPU and achieved a dramatic 5000X speedup with a 476040 atom viral capsid. More information at: http://people.cs.vt.edu/~onufriev/research.php

Computation of Atmospheric Composition (P.I.: Dr. Adrian Sandu)

Three-dimensional studies of atmospheric composition are computationally very intensive. Current model follows one hundred different chemical species in one million grid cells with time steps of hours for simulation time spans of years. These simulations are inherently data parallel. We have investigated the acceleration of chemical kinetic kernels on GPGPUs. The main challenge in real applications is the amount of data associated with the chemical kinetics in each grid cell. We have accelerated the STEM chemical transport model using CUDA on GPGPUs and have managed a speedup of about 5X. More information at: http://people.cs.vt.edu/~asandu/

Geoscience

Numerical Models of Mantle Convection on Earth and Terrestrial Planets. (P.I.: Dr. Scott King)

While motions in the interior of the Earth are almost imperceptible on human time scales, the process by which the Earth cools is the driving force behind most tectonic and volcanic activity. Finite Element Method (FEM) has been used to solve the equations of creeping convection. We are currently working with 24 Km grid resolution or 2.5 X 107 elements. With the help of Accelerators, we intend to reduce the time for current calculations from 20-30 hours to mere 30 minutes or less and make it possible to consider the real - time visualization. More information at: http://conman.geos.vt.edu/~sdk/#Research