Blacksburg, Va. — Virginia Tech researchers in computer science and
biology have used the university’s supercomputer, System X, to create
models and algorithms that make it possible to simulate the cell cycle
— the processes leading to cell division. They have demonstrated that
the new mathematical models and numerical algorithms provide powerful
tools for studying the complex processes going on inside living cells.
John Tyson, who studies the cell cycle, is a leader in applying
mathematical models in molecular cell biology. However, comparing the
results of a mathematical model to experimental data is difficult
because mathematical results are quantitative (numbers) while much
experimental data is qualitative (trends). The mathematical biologist
must figure out how to set the numerical values of the â€˜parametersâ€™
in the model equations in order to create an accurate representation of
what is going on inside the cell. A simple example is the conversion
between Fahrenheit and Celsius temperatures, said mathematician Layne
Watson. "You could use several pairs of Fahrenheit and Celsius readings
for the same temperature, and try to deduce the formula for converting
between the temperature scales."
Previously, Tyson worked with simpler models whose parameters could be
determined by trial and error, a process modelers call "parameter
twiddling." But he and his coworker, Kathy Chen, wanted to characterize
all the protein interactions regulating the cell cycle of budding yeast
(the yeast cells familiar to bakers and brewers, and a favorite
organism of molecular biologists, as well). "Such fundamental research
on the cell cycle of budding yeast provides a basis for understanding
the reproduction of human cells and is relevant to the causes and
treatment of cancer, to tissue regeneration, and to the control of many
pathogens," Tyson said.
For the budding yeast cell cycle, the
experimental data consists of observed traits of 130 mutant yeast
strains constructed by disabling and/or over-expressing the genes that
encode the proteins of the regulatory network. The model has 143
parameters that need to be estimated from the data. "That is a big
problem," said Watson. "You can’t do that by hand. You can’t even do it
on a laptop. It takes a supercomputer."
In fact, it required
more than 20,000 CPU hours on System X, a 2200 processor parallel
computer, using two new algorithms, DIRECT (DIviding RECTangles) and
MADS (Mesh Adaptive Direct Search), to estimate the 143 parameters.
"With a tool like this scientists can spend more time working on the model and less time twiddling parameters," said Tyson.
research is due to appear in 2007 in the Journal of Global
Optimization, in the article "Deterministic Parallel Global Parameter
Estimation for a Model of the Budding Yeast Cell Cycle," by Thomas D.
Panning, Layne T. Watson, Nicholas A. Allen, Katherine C. Chen,
Clifford A. Shaffer, and John J. Tyson.
Panning, who is from
Tulsa, Okla., received his master of science in computer science in May
2006 and is currently working as a programmer in Germantown, Md.
Watson, of Blacksburg, is professor of computer science in the College
of Engineering and professor of mathematics in the College of Science.
Allen, who is from Columbia, Md., received his Ph.D. in computer
science in November 2005 and is now with Microsoft. Chen, of
Blacksburg, is a research scientist biological sciences in the College
of Science. Shaffer, of Newport, is associate professor of computer
science. Tyson, of Blacksburg, is a University Distinguished Professor
of biological sciences.
The Virginia Tech computer science team
created massively parallel versions of a deterministic global search
algorithm, DIRECT, and a deterministic local search algorithm, MADS, to
do the twiddling, and then combined the results. "A deterministic
global search algorithm systematically explores the parameter space,
finding good values," Watson said. "Then the local search algorithm
improves the values from the starting points found by the global
The parallel computer programs can now be used by
others for similar problems. "The parameters found for the budding
yeast cell cycle model are good until the next scientist invalidates
them with new experimental data. That could be years from now or next
week. That’s the way science works," says Watson.
Source: Virginia Tech. January 2007.