< http://www.newscientist.com > New algorithms speed molecular simulations
17:35 21 August 01 Will Knight
Biologists and computer scientists have joined forces to create new algorithms that allow supercomputers to model molecular activity on an unprecedented scale. The technique could enable medical researchers to better predict the impact of drugs on cells "in silico", i.e. before any experiments on cells or animals.
The researchers, led by a team at the University of California, San Diego, used a recent mathematical discovery to accelerate hugely the speed at which supercomputers can process the data needed to simulate electrostatic atomic interactions.
This means that the number of atoms that can be modelled simultaneously has increased from around 10,000 to 1.2 million, allowing researchers to simulate biological activity at the level of molecules, not atoms.
Electrostatic models show how charges on individual atoms interact to produce electric fields throughout a molecule. This can determine the motion and stability of molecules and help biologists, for example, understand protein behaviour and the effectiveness of drugs.
"Electrostatic interactions are really important," says David Sept, a biologist at the University of Washington who contributed to the research. "We haven't had the means to tackle these problems until now."
Crunching numbers
A mathematical formula known as the Poisson-Boltzmann equation is vital to creating accurate electrostatic models. Calculating this equation conventionally requires a great deal of computer power, even for a simple model.
But mathematicians Mike Holst and Randolph Bank at the University of California discovered in 2000 that equations of this sort can be broken up into independent parts.
Nathan Baker, a postdoctoral researcher at the University of California, used this discovery to give different parts of the calculation to different processors in a supercomputer. This parallel processing produces significantly more computing power and allows far more complicated models to be created.
Mapping microtubules
The innovation has already yielded some useful results. The researchers used the technique to model the electrostatic charges on microtubules, which are part of cell's structure and transport system, as well as ribosomes, which are responsible for making protein within cells.
"Understanding microtubule instability and the mechanism by which they dissociate could have therapeutic applications, since many anti-cancer drugs act to stabilise microtubules," said J. Andrew McCammon, a chemist who led this research.
It took an IBM supercomputer, called Blue Horizon, less than an hour to perform the microtubule simulation. It ranks as about the eighth most powerful supercomputer in the world, with 1,152 different processors. The research team estimate that no computer could have performed this simulation using previous methods in a practical amount of time.
Journal reference: Proceedings of the National Academy of Sciences (vol 98, p 1037, p 10)
17:35 21 August 01