This page contains a Flash digital edition of a book.
biomedical modelling

from body/organ/cell investigations. If it were printed out, these fi ndings would typically run to thousands of pages. Consequently data handling and speed are vital features of these packages, exemplifi ed in the following supercomputing technique applied to blood analysis.

Blood vessel simulation Argonne’s Blue Gene/P supercomputer, housed at the Argonne Leadership Computing Facility (ALCF), allows scientists to tackle these immense problems with the power of 500 trillion calculations per second. One part of the study is mapping exactly

how red blood cells move through the brain. For example, last year the team used similar modelling to discover that the malaria parasite makes its victims’ red blood cells 50 times stiffer than normal. Healthy red blood cells are smooth and

elastic; they need to squeeze and bend through tiny capillaries to deliver blood to all areas of the brain. But malaria-infected cells stiffen and stick to the walls, creating blockages in arteries and vessels. Malaria victims die because their brain tissues are deprived of oxygen. A more complete picture of how blood moves through the brain would allow doctors to understand the progression of diseases that affect blood fl ow, like malaria, diabetes and HIV. This is an area of science called biophysics,

for the forces that govern red blood cells’ movements at this level are best described by the laws of physics and can be mapped with mathematics. That’s exactly what a team of scientists from Brown University led by G. E. Karniadakis and the US Department of Energy’s Argonne National Laboratory are doing on the lab’s supercomputer, hoping that a better map will lead to better diagnoses and treatments for patients with blood fl ow complications. Joe Insley, a principal software developer

at Argonne who is working with the team, comments: ‘Previous computer models haven’t been able to account accurately for, say, the motion of the blood cells bending or buckling as they ricochet off the walls. But this simulation is powerful enough to incorporate that extra level of detail.’ But before the simulations are even run,

there’s a hurdle that these researchers must face. It is a peculiarity of large computers that code for one computer doesn’t always work well on another. A code written for a computer with two cores – what’s probably in your home computer – doesn’t translate well into a computer that has 160,000 cores, as Argonne’s Blue Gene/P does. Because each supercomputer is individually

Qlucore Omics Explorer is used to visualise data is dynamic principal component analysis (PCA), an innovative way of combining PCA analysis with immediate user interaction

designed, the Blue Gene/P’s architecture is different from other supercomputers. Vitali Morozov, a computational scientist at the ALCF, explains: ‘For example, one of the Blue Gene/P’s strengths is good interconnects. The cores are beautifully arranged, and if you know how to use them it’s very effi cient – but it can be tricky.’ Thus, to get the best performance out of the machine, the code has to be tuned to fi t the computer.

Brain power In the area of research into neural activity, a number of recent neuroimaging studies have demonstrated the existence of a network of functionally connected brain regions that support a default mode of brain function. The work of George Zouridakis on modelling default brain connectivity networks at the Texas Learning and Computation Center in the University of Houston, Texas, depends on analysing large datasets of brain activity from in vivo analysis. Zouridakis has effected this by exploiting BLAS and scaLAPACK routines and OpenMP architecture. Granger Causality, a method for computing

brain connectivity networks, relies on autoregressive modelling of brain signals obtained using magnetoencephalography and electroencephalography. It can assess, for instance, the interaction between high-order executive cortical regions and low-level sensory and integration brain areas. Zouridakis comments: ‘We computed

Granger causality from resting state MEG, in an attempt to understand how the normal default network is affected by various neurological conditions, including, for example, mild traumatic brain injury and

autism. The ultimate objective of this line of research is to develop neurophysiological markers for various neurological disorders, such as TBI, post-traumatic stress disorder, and autism, to characterise disease accurately and assess the effectiveness of therapeutic interventions objectively.’ A major challenge in computing

connectivity network maps is the need to analyse large datasets consisting of hundreds of channels of brain activity obtained at high sampling rates. Depending on the spatial and temporal resolution of the original data, identifying the optimum model order for the brain signals and computation of Granger regressions may require several days of continuous processing on a desktop computer. To eliminate this bottleneck, Zouridakis’

team takes advantage of algorithm parallelisation and work distribution on high performance computing clusters that include hundreds of nodes by exploiting BLAS and scaLAPACK routines and OpenMP architecture. To optimise parallelisation, the parameters of the available nodes on the grid, such as shape and orientation of processors and block size during distribution of data among the processors, are selected empirically using fi ve benchmark neurophysiological signals of varying dimensions. He concludes: ‘The algorithms developed

are adaptive, so recordings of almost arbitrary size can now be analysed in hours instead of days.’

Optimising equipment design Besides looking into genetic and biophysical issues of the human body, biomedical simulation software is also critical for the design


Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12  |  Page 13  |  Page 14  |  Page 15  |  Page 16  |  Page 17  |  Page 18  |  Page 19  |  Page 20  |  Page 21  |  Page 22  |  Page 23  |  Page 24  |  Page 25  |  Page 26  |  Page 27  |  Page 28  |  Page 29  |  Page 30  |  Page 31  |  Page 32  |  Page 33  |  Page 34  |  Page 35  |  Page 36  |  Page 37  |  Page 38  |  Page 39  |  Page 40