IBM to tackle protein-folding with a special-purpose supercomputer, "Blue Gene", somewhat modeled on "Deep Blue" (pre-IBM name was "Deep Thought"), the computer that beat world chess champion Garry Kasparov.
This looks like the ultimate in "brute force" approaches to a complex problem. The computer will model the motion of the folding protein from first principles, recalculating everything in time slices of "a few quadrillionths of a second."
I'm not sure how much we'll learn about "consciousness, the origins of sex", but we're sure going to crunch some numbers.
Doc
p.s. The table below, which shows the surprising number of IBM gene-related patents, will also please INCY fans.
__________________
washingtonpost.com
IBM to Put Genetics on Fast Track
------------
_ Top Gene-Related Patent Holders _
Company/Organization Patents
U.S. Government 388
Incyte Pharmaceuticals 356
University of California 265
SmithKline Beecham 197
Genentech 175
Eli Lilly 145
Novo Nordisk 142
IBM 130
Chiron 129
American Home Products 117
(Source: PricewaterhouseCoopers)
--------------
By Justin Gillis
Washington Post Staff Writer
Saturday, June 3, 2000; Page A01
HAWTHORNE, N.Y. ?? Here is the plan: IBM scientists intend to spend five years building the fastest computer in the world, 500 times faster than anything in existence today. It will suck down every spare watt of electricity and throw off so much heat that engineers have bought a gas turbine the size of a jet engine to cool it.
The machine, dubbed Blue Gene, will be turned loose on a single problem. The computer will try to model the way a human protein folds into a particular shape that gives it unique biological properties. Obscure as it may sound, that kind of puzzle is at the heart of mankind's efforts to understand the nature of consciousness, the origins of sex, the causes of disease and many other mysteries.
Since proteins are the molecular workhorses of the human body, scientists would like to know the basic chemical rules for every one of them. Yet Blue Gene, 40 times faster than the combined speed of the 40 fastest computers in the world today, will run for an entire year to produce an answer for one protein.
That will be one down, 39,999 to go. Or thereabouts.
This is the biology of the 21st century. The project underway at International Business Machines Corp. is one notable effort to tackle a new problem, one so daunting it is already causing wrenching changes in science and industry.
Sometime later this year, scientists are expected to post a virtually complete human genetic sequence in the world's computer banks. Contrary to popular expectation, though, the completion of the human genome will be the end of nothing. It will, instead, be the beginning of a new race to unravel all the information encoded in genes. The real goal, likely to take decades to accomplish, is to understand in their entirety the ingredients and the chemical interactions that make up a human being.
"I am certain that a century from now, scientists will still be making major, insightful discoveries on the genetic sequence that's going to get determined this year," said J. Craig Venter, one of the scientists involved in the race to publish that sequence. "It's a mind-boggling amount of information."
Genes are scattered in bits and pieces along the 3.1 billion letters of genetic code that make up the genome. The genes are basically instruction sets for making proteins, and those proteins do nearly all the work associated with keeping a person alive. The kind of work that proteins do is determined by their three-dimensional shape, which is itself a function of their precise chemical makeup.
Proteins bind to and detach from other proteins to send messages, produce energy, store memories, eliminate waste and perform myriad other functions. These elaborate webs of cause and effect are referred to as biological pathways.
Which proteins bind to each other is largely a function of their shape. Mutations, or spelling errors, in the genetic code can lead to deformed proteins that don't work properly, and hence to major diseases. The particular pathways to which deformed proteins belong are said to be broken, and scientists want to find and fix them.
To do that, they need far better information about how the unique chemical composition of a protein, spelled out in the genetic code, leads it to snap into a particular three-dimensional shape as soon as it's created. That is the fundamental goal of IBM's program.
The company is paying for it, as it does many ambitious pure-science ventures, in hopes of being able to spin off new knowledge into products for the fast-growing market in biological computing. And research scientists at the company relish the chance to sink their teeth into a daunting problem such as how proteins fold up into working molecular machines.
"Nature does this day in and day out, second by second, and has done it for billions of years," said Sharon Nunes, a senior manager in computational biology at IBM. "We really want to understand the fundamental why and how."
As the IBM program illustrates, the scale of the problem that confronts biologists in the post-genome era is truly vast. Using the most modern techniques, teams of researchers can spend many years and millions of dollars working out the structure and function of a single protein. Now, thanks to genome research, they are about to learn of tens of thousands of new ones, many of them of urgent importance in cancer, heart disease and other ailments.
Biology seems to be going through the same kind of transition that hit physics half a century ago. In the early 20th century, individual geniuses such as Albert Einstein and Niels Bohr could make profound contributions to quantum physics using pencils. By the end of the century, further progress was being made mostly by scientists working in huge teams with billion-dollar atom smashers.
The most important event in 20th century biology came in 1953 when two young scientists, working with wire models and little encouragement, unraveled the basic chemical structure of deoxyribonucleic acid, or DNA, the carrier of hereditary information. Today, daunting sums--raised to a large extent on Wall Street--are being used to deploy robots, skilled technicians, acres of machinery and massive computers in the search for deeper answers.
The Human Genome Project, a government-backed consortium of laboratories across the globe, is expected to produce a draft gene map next month and a final map by 2003. By combining its own data with those of the public consortium, Celera Genomics Corp. of Rockville is expected to produce what it describes as a virtually final map by this fall.
These and similar, though less ambitious, initiatives all aim toward the same goal: To come up with an ingredient list for the human body. Hard as it may be to believe, all the medical progress of the last century has come without biologists having a list of human proteins, much less a clear fix on how they interact.
"When you go to get your car fixed, you take for granted that the mechanics know what all the parts are. They're not discovering parts on the job," said Eric Lander of the Whitehead Institute for Biomedical Research in Cambridge, Mass., a leader in genome research. "Surely, for something as important as your body, one should be appalled that we've been trying to practice medicine--trying to practice car repair--without even knowing all the parts."
The completed gene maps will be vital tools, but they will not automatically give scientists adequate information about the proteins the genes encode or how they fit together.
"Having the parts list is only the beginning," Lander said. "If I gave you the list now for a Boeing 777, which has over 100,000 parts, that list would not tell you how to put it together and it would not tell you why it flies."
IBM's new computer represents one approach to these deeper biological questions. It is controversial among researchers. Some of them tend to think that IBM has chosen an excessively difficult, or at least premature, method. The company's scientists reply that they think their approach will work.
IBM's ultimate goal is not just to model one protein, but to come up with some general rules about how proteins behave--not only how they fold but what shapes are likely to bind to what other shapes.
To do this, IBM scientists intend to cut a protein's formative stage into trillions of individual slices of time. Then, starting with the elementary principles of chemistry and physics, they will tell Blue Gene to calculate the action of atomic forces at each time point.
This modeling is likely to work only if researchers use extremely short time slices, each a few quadrillionths of a second long. That means the same complex calculations have to be performed over and over, which is what makes the problem so complicated and time-consuming.
To many biologists, awed though they are by IBM's computer prowess, this approach falls into the category of "too much information." They note that proteins fold into the correct shape inside cells in the blink of an eye. Many researchers are content to start from there and not worry about the elementary chemical principles underlying that shape. Moreover, they already have experience finding dysfunctional proteins by trial and error.
In a sense, they want to apply shortcuts to the study of proteins as a way of producing important information in a hurry. In the long run, it's likely researchers will want the kind of detailed chemical information that will come from projects like IBM's, but in the short run, some of them are betting they can get important leads about cancer, heart disease or other ailments more quickly.
But even with the shortest of shortcuts, the key question--which proteins, out of all those in the human body, bind to one another, and for what purposes--is still daunting. As gene-mapping efforts near completion, companies and universities are suddenly racing to catch this wave. New fields with names like "functional genomics" and "proteomics" are being developed to study, on a massive scale, the ways that genes and proteins work in the body.
Under any scenario, though, these questions are going to be answered only with a massive deployment of robots, computer power and industrial-scale research techniques.
Researchers at one biotechnology company, CuraGen Corp. of New Haven, Conn., in cooperation with the University of Washington, awed biologists in February by publishing, as a cover article in the journal Nature, a paper describing the interactions of all known proteins in brewer's yeast, Saccharomyces cerevisiae, a vital study organism. It was the first time anyone had done a "global" survey of protein interactions for a complex type of cell.
The company plans to apply its techniques to the study of human proteins, and it will have plenty of competition. CuraGen's chief executive, Jonathan Rothberg, said even partial information about biological pathways can be useful.
"I only want enough information to make a drug," Rothberg said. "Our goal is to get to the point that we can make drugs that are effective against complex genetic diseases that have never been cured before. If I make a drug for diabetes or an antibody that works against cancer, I've done my job."
THE BIGGEST BLUE
Blue Gene, the computer IBM is building to analyze protein folding, will be:
40X
About 40 times more powerful than the aggregate power of the 40 fastest supercomputers in the world today.
500X
About 500 times more powerful than the fastest computer in the world today, a machine called ASCI Red at Sandia National Laboratories in Albuquerque.
1,000X
About a thousand times more powerful than Deep Blue, the computer that beat Gary Kasparov in chess.
2,000,000X
About 2 million times more powerful than the fastest desktop computer available today.
SOURCE: International Business Machines Corp.
¸ 2000 The Washington Post Company |
