washington

Complex number: this computer image generated at the National Center for Supercomputing Applications at the University of Illinois shows a model ion channel in a lipid bilayer embedded in a solvent milieu. The image demonstrates the complexity of macromolecular systems. The centre is funded by the National Science Foundation. Credit: SHANKAR SUBRAMANIAM/NCSA

A group of biologists who use supercomputers to model the structure and mechanisms of biological molecules is urging the US National Institutes of Health (NIH) to boost its investment in high-performance computing.

The group, which gathered last week at a meeting in Rockville, Maryland, sponsored by NIH's National Center for Research Resources (NCRR), was virtually unanimous in arguing that NIH is lagging behind leading federal agencies — namely, the Department of Energy and the National Science Foundation (NSF) — in investing in supercomputers.

The suggestions of the group. which will be given to NCRR, include the recommendation that NIH fund a single, central facility that operates a large ‘teraflop’ machine, the next generation of supercomputer.

The biologists said that the next generation of ‘teraflop machines’ — a hundred times more powerful than existing supercomputers — would give qualitative gains, including the possibility of modelling protein-protein recognition and self-assembly. They also spoke of the possibility of studying larger molecular systems, rather than, say, isolated enzymes, leading to a more sophisticated approach to drug discovery.

They also say the machines would allow far longer simulations. As it stands, studies of the changing forces within biological molecules are limited to nanosecond glimpses, preventing tests of theories of molecular structure and function in the real world, where changes take a million times longer.

But “NIH for some reason is just not there,” says J. Andrew McCammon, professor of pharmacology and chemistry at the University of California, San Diego. He uses the San Diego Supercomputer Center, one of two major centres funded by the NSF.

“The NIH are not a player in this high-performance computing business, and they really need to be,” said Klaus Schulten, who chaired the meeting and heads an NCRR-funded mid-level computer resource at the University of Illinois at Urbana-Champaign. He added that NIH was being asked “with a louder and louder voice” to fund computing centres, but that the biomedical agency has “not addressed the role of computing⃛ as forcefully as some other agencies”.

The meeting was organized to list the advances that teraflop machines would allow, to present NIH with a convincing argument for investing the multi-million dollar sums required to make high-performance machines more broadly available.

The NIH supports only one $3.9 million, dated supercomputer centre at a National Cancer Institute campus at Frederick, Maryland. Biologists do most of their work at a centre in Pittsburgh, mainly funded by the DoE, and at the NSF-funded centres at San Diego and Urbana.

NSF invests $70 million annually in its two centres. And the $366 million government-wide computing initiative recently announced by Vice-President Al Gore allocates only $6 million to NIH, of which just $2 million goes to “advanced computing” (see Nature 397, 285; 1999).

The use of the NSF centres by biologists is growing rapidly. Between June 1997 and May 1998, one-third of the time at the NSF supercomputer centres was allocated to (mostly molecular) biologists. This is a 54 per cent increase on the previous year.

“There's a very clear trend that the major usage and the large projects are swinging over from being physics- and astronomy-driven to being biology-driven,” says Eric Jakobsson, a senior research scientist at the NSF-funded National Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign. “In a couple of years the cycles for big users will probably be about 50 per cent biology.”

But this does not satisfy biologists, who feel as squeezed as the physicists and astronomers with whom they share the machines, and say that ‘cycle drought’ — the chronic lack of computer time — should compel NIH to act on their behalf. Overall, requests for time on the leading computer at San Diego exceed that available by a factor of four or five.

Harold Varmus, the NIH director, commissioned a working group on biomedical computing last year to produce a report that will be presented to his advisory committee in June. Part of the group's remit is to investigate “the impediments biologists face in utilizing high-end computing”.

Larry Smarr, director of the NSF centre at the University of Illinois and a co-chair of the working group, says he agrees that supercomputers offer a “tremendous opportunity” to biologists. But he says there is a lively debate in his group on how to increase NIH-funded high-performance computing, and no consensus has been reached. The group is looking at the extent to which enlarging existing ‘clusters’ of personal computers or workstations in research groups could meet the need for advanced computer power.

Also, should NIH invest in a new supercomputer centre, build on the one at Frederick, or develop several smaller centres dedicated to applications such as organ simulations, genomics and the study of biopolymer systems? Or should it add its resources to one of the NSF or energy department centres, where it could profit from cross-pollination with the physical sciences and the existing infrastructure? “My guess is that there will be some sort of combination of solutions,” says Smarr.

But, he says, whatever his group recommends, the supercomputer needs of biologists will not be assessed in a vacuum. “Clearly the science that's done with supercomputers can't be done any other way,” says Smarr. “But there are vastly more biologists that use the web than use supercomputers. If [NIH] has only so many dollars, then it has to make decisions across the spectrum.”

Many of those at last week's meeting argued that the kinds of advances foreseen by advocates of teraflop machines do not fire the imaginations of ordinary biologists. They pointed out that, to sell high-performance computing to NIH, enthusiasts will have to chart a course from these improvements to tangible health benefits.

That's not difficult, says McCammon. He points out that the HIV protease inhibitors now prolonging thousands of lives were developed in part with computational methods from his laboratory.

He says that his lab's work on the neuroenzyme acetylcholinesterase, discovering computationally how the enzyme gets its speed and specificity of action, is laying the groundwork for better drugs for neurological disorders from Alzheimer's disease to glaucoma.

Others said that, regardless of the power of supercomputers or NIH investment, their availability will be partly wasted unless the broader community of biologists becomes better educated in their use.

There is “huge ignorance” among ordinary biologists about how to use computational tools, said Carol Post of Purdue University, Indiana. “The universal problem is not hardware,” adds Smarr. “All the biologists [the working group has interviewed] say ‘we need more training’.”