Characteristics of eScience
- High Performance Computation
- Distributed infrastructure
- Instruments are first class resources
- Lots of data
- Not just bigger fundamentally different
Overview of Message Lab role in eScience
One of the defining features of research conducted in Message Lab is that it is applications driven - that means that we use real applications to push the envelope of the tools and techniques that we develop. One outcome of this is that we have assembled a wide range of applications across different domains, and in many cases performed in collaboration with researchers from across the globe. This page shows a selection of eScience projects that have employed our tools and demonstrates their wide applicability.
eScience Research Themes
Message Lab's involvement in eScience follows a number of themes or disciplines, namely:
- Chemistry and Physics
- Medical and Life Sciences
- Engineering and Design
- Mathematics and Computer Science
- Economics and Finance
- Environmental Science
- Earth Sciences and Astronomy
Chemistry and Physics
with Wibke Sudholt, Kim K. Baldridge, University of Zurich
Quantum mechanical computer chemical models give accurate predictions but are computationally impractical for large molecules. Hybrid models use quantum mechanics to describe a small active region, and classical molecular mechanics for the rest. Combining the two methods is difficult. This work investigated the pseudobond approach where a hypothetical capping atom is used to saturate the chemical bonds.
Optical Signal Processing
with Malin Premaratne, Monash University, and Bipin Pillai, University of Melbourne
Semiconductor optical amplifiers provide amplification of optical signals without needing to convert to an electrical signal. This work obtains analytical expressions for properties of an optical pulse propagating in such a device, expressions that are useful in optical signal processing applications.
with Alan Bond, Chongyong Lee, Elena Mashkina, Monash University, and David Gavaghan, University of Oxford
Voltammetry explores the electrochemical reactivity of an analyte by inserting that analyte into an electrochemical cell (in solution or attached to an electrode) varying the applied voltage and measuring the resulting current. By comparing the performance of the real system with computer simulations, estimates of analyte properties may be obtained. This is a standard inverse computational problem, finding model parameters that yield outputs matching expected results. Nimrod/O is a convenient tool for such work.
with Matthew Dimmock, John Gillam and Toby Beveridge, Monash University
Emission imaging maps the distribution of radio-tracers injected into a patient as they spread through the blood stream. By selecting a tracer that accumulates at certain anatomical features, the presence and location of abnormal uptake e.g. tumors, can be quantified. This is achieved through the detection of gamma-rays emitted from the radio-tracer.
PEDRO is a new prototype imaging system. The design involved a choice of the geometry of the detectors in order to optimize the signal to noise ratio of the reconstructed image. To achieve this, a mathematical model of the system was built and run under Nimrod/G to explore the design space, and also under Nimrod/O to determine optimality.
Medical and Life Sciences
with Kim K. Baldridge, University of Zurich
Docking of drugs with cells involves interaction between relevant proteins and ligands. Models of this process use either classical molecular dynamics or quantum mechanics. The latter is more accurate but computationally infeasible with complex organic models.
This project has developed a hybrid algorithm, sufficiently accurate and computationally practical, that computes the binding energy for a given protein-ligand geometry. Nimrod/O was used to optimize that binding energy. The strength of the binding gives a measure of the efficacy of the drug.
Electrical activity of the heart is a product of that of the component myocyte muscle cells which in turn is produced by the flow of ions through ion channels in the myocyte membranes. More detailed models of ion flow processes are being developed with the ultimate aim of understanding both normal and dysfunctional hearts.
Ion Flux in Cardiac Models
with Andrew McCulloch, Anushka Michailova and Roy Kerchoffs, University of California, San Diego
This project used a new model of rabbit ventricular myocytes, running under Nimrod/O, to determine settings for the metabolic factors that would produce realistic behaviour.
Improved Models of Calcium Ion Dynamics
with Anna Sher, David Gavaghan, Denis Noble, and Penelope Noble, Oxford University
This work aims to improve existing ventricular cell models by replacing their description of Ca2+ dynamics with the local Ca2+ control models. This required that the parameters of the Ca2+ subsystem be re-fitted. Nimrod/O was used to optimise these.
The Effect of Ion Channel Conductance on the ECG
with Miguel Bernabeu, Alberto Corrias and David Gavaghan, Oxford University
Variations in the conductance of the potassium ion channels of cardiac cells are known to significantly affect the electrical performance of the heart. Such conductance may be congenitally impaired, or affected by drugs (either deliberately or as a side effect of other medication). This work used computational models to investigate the effect on the electrocardiogram (ECG).
with Mary Chin, Velindra Cancer Centre, U. K. and Jonathon Giddy, Cardiff University U. K.
The use of computer models of radiotherapy predicts the dosage delivered to various parts of the body. In a clinical situation this enables a regime tailored for the needs of each patient, to maximise the dose to a tumour while minimising the dose to healthy tissue.
To achieve the required accuracy such models are computationally demanding. This project used Nimrod to distribute the jobs.
Determination of Protein Structure
with Ashley Buckle, Monash University
Structural biology research places significant demands upon computing and informatics infrastructure. Protein production, crystallization and X-ray data collection require solutions to data management, annotation, target tracking and remote experiment monitoring. Structure elucidation is computationally demanding and requires user-friendly interfaces to high-performance computing resources. We are developing Grid computing approaches, using NIMROD, that aim to automate the structure determination process by allowing hundreds to thousands of candidate models to be tested in parallel, along with parameter sweeps. This infrastructure will be pivotal for accelerating the process of structural discovery.
Parametric Studies of Gamma Oscillations
with Evan A. Thomas, Jordan D. Chambers, and Steve Petrou, University of Melbourne
Gamma oscillations in the brain are thought to be critical for a number of behavioural functions, even conscious awareness. A better understanding of the mechanisms involved could assist in treatment of epilepsy, for example. Complex computer models of the cortex exhibit oscillations in the frequency range of gamma oscillations, so may provide insight into the real system. The model used involved 33 parameters that are relevant to the power and frequency of these oscillations. This is too many for a sweep of the entire parameter space,so Nimrod/E was used to generate an appropriate experimental design.
Engineering and Design
with Clive Fletcher, CANCES, University of New South Wales, and Andrew Lewis, Griffith University
Computational fluid mechanics provides a mechanism for assessment of the aerodynamics of engineering designs. This project modelled the airflow around a wing profile to determine how the ratio of lift to drag varies with certain shape parameters. Nimrod/G was used to survey the parameter space and Nimrod/O to determine the shape that gives optimal lift to drag.
Flame Kernel Growth in Turbulent Flows
with Karl Jenkins , Cranfield University, UK
Efficient combustion in an engine requires a high proportion of the fuel to be burnt, which in turn depends on the flow of the fuel/air mixture. Computational models are now capable of predicting the turbulent flows involved
This research built new models of the turbulent flow and propagation of the flame front. Nimrod was used to sweep over values of certain parameters in order to determine optimal efficiency.
Human-Computer Interaction Design for Air Traffic Control
with Andrew Neal, Jacqueline Wicks and Peter Lindsay,
ARC Centre for Complex Systems.
This project is developing a new approach to human reliability assessment of human-computer interaction design options with application to air traffic control. Operator Choice Models were developed for conflict detection and resolution tasks using experimental data collected from simulator trials with student subjects. The formal models were further used to explore the effect on performance of different computer-based tools. A parameter sweep was conducted, using the Nimrod tool, to compare predicted operator performance for four different design options on a range of different traffic patterns.
with Andrew Lewis and Seppo Saario, Griffith University
The design of a test rig for a handset antenna incorporated a ceramic bead in order to reduce the distortion of the radiation pattern during testing. The design needed to choose properties of the bead to minimise that distortion. This was calculated using an electromagnetic model of the antenna and rig. As each calculation required over 20 minutes on a high end workstation, Nimrod was used both to explore the design space for the bead properties, and to find optimal settings.
Fatigue Based Design Optimization
with Rhys Jones, Kumanan Krishnapillaia and Daren Peng, Monash University
Fatigue failure in structures is caused by the growth of cracks when subjected to repeated stresses. The process can now be modelled using finite element analysis to compute the stress field and models predicting the growth of crack tips within that field. Nimrod/G has been used to determine the effect of controlable parameters and Nimrod/O to determine optimal shapes that minimise crack growth.
Optimization of Renewable Resources in Electricity Grids
with Ariel Liebman, Leighton Brough and Liam Wagner, Centre for Complex Systems, The University of Queensland
Plexos proprietory software is used to model the Australian electricity distribution network, in particular, to study the inclusion of renewable resources. The principal aim is to answer the question "what is the optimal allocation of embedded renewable generators in the distribution network which minimises the expansion cost of the distribution/transmission system and the price of carbon credits?"
Multi-objective Shape Optimization of Structural Engineering Designs
with Mike J. W. Riley, and Karl W. Jenkins, Cranfield University, UK
Optimization of engineering design typically involves a trade-off between competing requirements. In this research into the design of a rib-reinforced wall bracket, the desire for minimal strain and stress conflicted with the requirement of minimal mass. In such problems the designer may seek Pareto points, designs for which one objective can only be improved at the expense of another. Here an evolutionary multi-objective optimization code, DEMO, was interfaced with Nimrod/O and with the finite element package Code_Aster to delineate the Pareto points.
Mathematics and Computer Science
Recommender Systems for Museum Visitors
with Fabian Bohnert and Ingrid Zukerman, Monash University
Recommender systems process information about a person with the aim of predicting items of interest to that person. This project uses observations of the movement of visitors in a museum to recommend interesting exhibits, using a Spatial Process Model. The model used the Gibbs Sampler to estimate the required parameters. Such a method is computationally intensive; for this work requiring over 1900 hours on high-end processors. Nimrod/G was used to control the experiment, farming work to both a Monash cluster and to cloud resources.
The Norm of the Generalized Stieltjes Transform
Tom Peachey and Colin Enticott, Message Lab, Monash University
Integral transforms are commonly used in Physics and Engineering. The generalized Stieltjes transform, for example, has been applied in string theory. But the "norm" of this transform (its effect on the "size" of the functions transformed) is a long unsolved mathematical problem.
This project developed software to evaluate the effect of the transform on a particular class of functions. Then the evaluation of the norm becomes an optimization problem on the output. Nimrod/O was used to perform the optimization, running inside a Nimrod/G sweep which varied the relevant parameters.
Economics and Finance
Estimating Complex Production Functions
with Mark Neal, University of Queensland
Production functions are rules that relate economic outputs to the necessary inputs. This work used maximum likelihood methods of estimating the coefficients in these rules. The computations used Shazam econometric software running under Windows. By distributing the task using enFuzion (the commercial version of Nimrod), the computational task was reduced from several weeks to a few days.
Conditional Duration Models
Kulan Ranasinghe and Mervyn Silvapulle, Monash University
A major research effort in Econometrics is the development of new tools for estimating population parameters from statistical data. This research concerns the estimation of the distribution of elapsed time between consecutive events, for example the times between trades on a stock exchange. A new method is proposed and the efficacy and accuracy explored by extensive simulations using a Matlab model running under Nimrod/G.
Sensitivity of the Australian Monsoon to Savannah Fire
with Amanda Lynch, Klaus Görgen, Monash University
This project investigates the climatic effects of a change in the fire regime in the savannah lands of northern Australia. Each simulation spanned a 21 year frame; 90 simulations were required, generating a major computational task. This was executed using Nimrod/G.
Climate Modelling with Web Services
with John McGregor and Jack Katzfey, CSIRO Marine and Atmospheric Research, Melbourne.
The task of computing a global climate model at a fine enough resolution to capture local features is computationally huge, and near the limit of current technology. Instead a global model (the Conformal-Cubic Atmospheric Model) simulation is computed with a coarse grid and the results used to drive a simulation having a fine grid over the local region of interest.
For a variety of reasons, the two models may be performed at different geographical locations. This project demonstrated the use of web services to control the workflow. In particular, the GriddLes
approach enabled the models to communicate without modification and even allowed the local modelling to begin before the global model was complete.
Earth Sciences and Astronomy
Interactive Inversion in Structural Geology
with Louis Moresi and David May, Monash University, and Dietmar Müller, University of Sydney
Geophysical models of evolution of the earth's crust require specification of various model parameters. This project was concerned with the inverse problem of determining input parameters that produce realistic earth dynamics. This is can be phrased as an optimization problem, minimizing the difference between the model output and realistic evolution. However, assessment of such differences is best achieved by the subjective evaluation of expert observers. For this purpose, Nimrod/O was extend (to Nimrod/OI) to allow interactive assessments of the animations produced by the model.
with Kurt Liffman, CSIRO Materials Science and Engineering, Melbourne
Meteorites are assumed to have formed in the solar nebula during the formation of the solar system. This work investigates the hypothesis that they formed in the innermost regions of the nebula and were then ejected to outer regions. Such a history should have effects on the size sorting of the chondrules and metal grains that compose the meteorite.
This project performed Monte Carlo simulations of this formation process, using Nimrod to distribute the work.
Volcanic Ash Deposition
with Santiago Núñez and colleagues, Costa Rica Institute of Technology
Prediction of the path of volcanic ash fallout is not just of interest to the airline industry. In the long run, such ash produces fertile soils. The people who settle to farm these soils are at risk from later eruptions; ash can directly damage crops and structures and the acid it carries can cause respiratory disease.
San Jose, the capital of Costa Rica, suffered recurrent major ash deposition over the years 1963-5, from the Irazu volcano. This research applies the Suzuki model of ash deposition to that eruption. A new tool NG-TEPHRA, which implements the Suzuki model has been developed to run in parallel under Nimrod/G. The model is parallelised across a location grid and requires many input parameters that specify the intensity of the eruption, size and shape of the ash particles, and the wind strength and direction at various elevations.
Recent work (to appear in proceedings of IEEE e-Science 2010) focused on development, testing and quantification of an early parameter-verification mechanism used to discard unwanted sets prior to allocating remote resources. This is useful for applications such as NG-TEPHRA, which exhibit non-homogeneous parameter spaces due to parameter inter-dependency and/or employ selective pruning for optimisation.