Main menu


HPCSYM13 Program

Download a final program or poster for printing.

Thursday, April 11, 2013

Lehigh Seminars: Linderman Library, Bayer Galleria (Third floor)

Time Event
1:30 PM - 2:30 PM
Michael Chupa
Research Computing at Lehigh and Beyond
2:30 PM - 3:00 PM
Coffee & snack break
3:00 PM - 4:00 PM
Stephen Anthony
Efficient Use of Lehigh HPC Resources

Friday, April 12, 2013

Symposium technical program, Linderman Library 200

Time Event
8:00 AM - 8:45 AM
Registration / coffee
8:45 AM - 9:00 AM
Welcome and Introduction
Bruce Taggart, Vice Provost for Library and Technology Services, Lehigh University
9:00 AM - 9:50 AM
Jing He: Deriving Protein Tertiary Structures from Electron Cryo-microscopy Volume Data—the Computational Challenges
10:00 AM - 10:50 AM
Avner Schlessinger: Drug discovery by homology modeling and virtual screening: application to human Solute Carrier (SLC) transporters
10:50 AM - 11:00 AM
11:00 AM - 11:50 AM
D. John Hillier: A Spectrum is Worth a Thousand Pictures
12:00 PM - 1:00 PM
Luncheon, Linderman 200
1:10 PM - 2:00 PM
Keynote speaker Ward Wheeler
Heuristic algorithms, optimality criteria, and computational challenges for sequence data: DNA to Human Language
2:10 PM - 3:00 PM
Kostas Bekris: Progress and Prospects in Motion Planning in the Era of Cloud Robotics
3:15 PM - 3:30 PM
coffee / snack break
3:30 PM - 4:20 PM
Hubertus Franke: Software Defined Environments

Keynote Lecture: Ward Wheeler

Chair, Division of Invertebrate Zoology; Curator-in-Charge, Scientific Computing; Curator, Invertebrate Zoology, American Museum of Natural History

Heuristic algorithms, optimality criteria, and computational challenges for sequence data: DNA to Human Language

Many types of comparative data come in the form of sequences—linearly ordered sets of elements. These sequences undergo a variety of transformations that can be used to infer phylogenetic relationships. Unfortunately, many of the problems encountered in the phylogenetic analysis of sequences are NP-Hard, requiring heuristic solutions. In this talk, several computational approaches and optimality criteria will be applied to the historical analysis of DNA, protein, and human language data sets.

Kostas Bekris

Computer Science Department, Rutgers University

Progress and Prospects in Motion Planning in the Era of Cloud Robotics

Some of the fundamental challenges in robotics and simulation lie in the area of motion planning, which studies algorithms that return actions for safely accomplishing a physical task. Interesting problems include path quality guarantees, effective planning for systems with significant dynamics, real-time considerations and multi-robot coordination issues. This talk will review the motivation for new motion planning solutions and present recent work regarding (a) asymptotic and finite-time near-optimality, including for systems with dynamics, (b) and multi-robot challenges, ranging from complete multi-agent path-finding to decentralized motion coordination. The talk will conclude with the long-term prospects of this line of research and the opportunities it provides in the application domains especially in the context of the emerging concept of cloud robotics, the integration of cloud computing with robotics.

Hubertus Franke

IBM Research Division, Thomas J. Watson Research Center

Software Defined Environments

A new model of computing, we call Software Defined Environments (SDE), is emerging. In software-defined environments the system infrastructure as well as the software can be flexibly tuned towards the needs of individual workloads. Workload requirements are captured at an abstract level through DevOps and during deployment the requirements are translated into the necessary software and infrastructure resources, which are then provisioned on demand and continuously optimized for improved outcome. This improves the overall agility, efficiency and operational stability of systems.

Jing He

Computer Science Department, Old Dominion University

Deriving Protein Tertiary Structures from Electron Cryo-microscopy Volume Data—the Computational Challenges

Electron cryo-microscopy is becoming a powerful technique to derive the near atomic structures of large molecular complexes such as viruses, membrane protein complexes, cytoskeleton fibers and ribosomes. It is still computationally challenging to interpret the data generated from this technique. We have been taking a multidisciplinary approach to derive the near atomic structures of proteins from experimental image data at medium resolutions between 5 and 10 Å. In order to derive the protein structure, we have used image processing and computational geometry techniques to detect patterns from 3-dimensional volume data, and we have developed effective graph algorithms to derive the topology of the secondary structure elements of the protein. The major computationally intensive steps involve the modeling of the protein chain and the determination of the topologies when the errors in the data are considered. Effective parallel algorithms are needed.

D. John Hillier

Dept. of Physics and Astronomy, Univ. of Pittsburgh

A Spectrum is Worth a Thousand Pictures

Much of what we have learnt about nebulae, stars, and galaxies comes from studies of their spectra. Through spectra we can, for example, determine the temperatures, mass, radii, and compositions of astrophysical objects. In this talk we highlight the difficulties of accurately modeling spectra, particularly for objects such as supernovae. To model spectra accurately we need to model the transfer of radiation through the emitting medium. To do this we often make simplifying assumptions, such as spherical geometry. In general, the transfer of radiation is six-dimensional—three spatial dimensions, two angular dimensions, and time. The high dimensionality makes modeling radiation transfer computationally demanding.

The computation of spectra is often further complicated by the low densities often encountered. As a consequence of the low densities, we cannot assume local thermodynamic equilibrium—that is, we cannot assume the atomic level populations are in thermodynamic equilibrium, and hence only determined by the local density and temperature. Instead we must solve for the populations of each atomic level. This is complicated since there are thousands of atomic levels, the level populations and radiation field are coupled (the level populations depend on the radiation field which in turn depends on the level populations), and we need to solve for the level populations at each “point” in the medium.

We highlight progress in modeling supernovae using our time-dependent radiation transport code. Supernovae are among the most energetic objects in the Universe. But supernovae are more than just energetic events—they are essential for life. By studying supernovae light curves and spectra we gain insights into stellar evolution, the nature of the progenitor star, surface abundances at the time of the explosion, whether previous mass-loss episodes have occurred, and the physics of the explosion.

Avner Schlessinger

Department of Pharmacology and Systems Therapeutics, and Tisch Cancer Institute, Mount Sinai School of Medicine

Drug discovery by homology modeling and virtual screening: application to human Solute Carrier (SLC) transporters

Solute Carrier (SLC) transporters are membrane proteins that control the uptake and efflux of a broad spectrum of substrates, such as nutrients, toxins, and prescription drugs. In human, there are 386 SLC transporters that can be drug targets themselves or be responsible for absorption, disposition, metabolism, and excretion of drugs. A recent increase in the number of atomic structures of several SLC homologs combined with technological advances in structural modeling and ligand docking has expanded our ability to characterize many previously unmodelable SLC transporters. Here, we first perform a comprehensive comparison of the SLC transporters to inform attempts to model their atomic structures, a prerequisite for structure-based ligand discovery. We then describe an integrated computational and experimental approach for identifying transporter-small molecule interactions. Particularly, we use homology modeling and virtual screening, followed by experimental validation by measuring uptake kinetics, to identify interactions between SLC transporters and small molecules ligands, including prescription drugs, metabolites, and fragment-like compounds. For example, we discovered that several existing prescription drugs interact with the norepinephrine transporter (NET), which may explain some of the pharmacological effects (i.e., efficacy and/or side effects) of these drugs. Our combined theoretical and experimental approach is generally applicable to structural characterization of protein families other than transporters, including receptors, ion-channels, and enzymes, as well as their interactions with small molecule ligands.