Posted on 09/06/2005 3:04:35 PM PDT by Ernest_at_the_Beach
By Alex Lash
Story last modified Tue Sep 06 09:00:00 PDT 2005
Engineers build bridges knowing exactly how every piece fits and how every physical force interacts, but that's often not the case with pharmaceuticals.
Much of what happens on the molecular pathway to developing drugs remains murky. Bridges are obviously less complex than people, but there are ways to take some of the guesswork out of drug discovery. "People simulate spacecraft and bridges," said Stanford University professor Vijay Pande. "Why not simulate ourselves?"
That's a tall order, of course, yet as computing power increases, life science researchers such as Pande can simulate more of what actually happens when disease attacks a body and, in turn, what happens when a drug attacks the disease. That's good news for the pharmaceutical industry, which desperately needs to cut its costs and improve its odds of developing useful drugs. And with the data the genomic research boom has unleashed, drug companies big and small are transferring more of their wet lab work in vivo to virtual computing environments in silico.
The latest leap in supercomputing power comes from IBM's Blue Gene machines. The most powerful supercomputer, installed at the U.S. government's Lawrence Livermore National Laboratory, can perform more than 100 teraflops, with higher speeds coming later this year. According to the Top500 Supercomputer list, Blue Gene machines now occupy the top two spots, and five of the top 10 fastest computers in the world.
The name of IBM's latest systems is no accident. Big Blue is mounting a full-court press to be the hardware supplier of the life sciences industry, which is getting more computing-intensive by the day. The turning point in the relationship between computing power and life science advances came with the efforts to crack genomic codes. With genomic research now spilling forth vast volumes of data, ever more computing power is needed to peer deeper into how diseases work and find cheaper, faster and safer treatments.
"It's a big game changer," says Ajay Royyuru, a computational biologist in charge of IBM's life science research. "We now have the ability to think about the whole system, not just parts. We can start to build a robust model of how a system is behaving or misbehaving and ask, 'How can I correct it?' If I knock a gene out, how will the whole system behave?"
The ability to simulate how an entire system behaves is the modus operandi of Blue Gene, which ties the brute strength of tens of thousands of processors with a sophisticated architecture that lets processors share information on the fly. The record-breaking versions are running in elite institutions such as the Lawrence Livermore Lab--its behemoth consists of 32 e-Servers lashed together--and Japan's Advanced Institute of Science and Technology, but the basic 5.7-teraflop "e-Server" system is available for roughly $2 million. IBM is also renting space on Blue Gene systems for about $10,000 a week.
It ain't cheap
For the top researchers, the great leap forward in computing is worth it. But much drug discovery work doesn't require Blue Gene-like sophistication. "With Blue Gene, you pay a premium for the on-the-fly communication," said Colin Hill, chief executive of Gene Network Sciences, which uses computer models to help drug companies figure out if a drug will have any effect, beneficial or toxic, on a virtual patient.
To do so, GNS needs lots of brute strength to crunch through billions of nearly infinite scenarios. Because such work is often done on many processors at once, the more computing power, the better. GNS's largest computer is an Apple server, "but we're limited by the power we can afford," Hill said. "We just don't have $20 million to throw at it."
Another Blue Gene customer also worries about the cost. Quantum Bio crunches data to help big pharmaceutical companies find potential drugs. The company has just ported its software to Blue Gene and wants to give its customers remote access, but IBM's $10,000-a-week price tag is a tough sell, said chief software architect Lance Westerhoff. "Now that everyone wants pharmaceuticals cheap, they have to cut costs," Westerhoff said.
But in these early days of computational drug development, Westerhoff says the cost-benefit ratios are hard to pin down. If clinical trials--where drugs are tested on humans--could somehow be simulated, costs could shrink dramatically. But clinical trials are the least computerized step in the drug-approval process. There's no substitute for giving drugs to patients and watching what happens.
For now, computers are more effective in preclinical discovery,
eliminating bad drug candidates early in the process. But bad ideas that get nipped in the bud are a hard cost to quantify, and there's no guarantee one won't slip through and waste enormous time and money.
On the other coast, Stanford's Pande is using supercomputing power to explore one of the great frontiers of molecular biology. His Folding@Home project simulates how proteins are formed: Long chains of amino acids, cued by genetic instructions, accumulate and fold into precise 3D structures. Scientists believe protein misfolding may be key to diseases such as Alzheimer's, Parkinson's and some cancers.
Pande's project isn't using Blue Gene hardware, although they collaborate on some software. His team has linked PCs around the world that work on tiny pieces of the puzzle during their idle time. This "distributed" computer with the power of 100,000 processors is in some ways faster than Blue Gene. After four years, the project is close to simulating a protein folding for as long as a millisecond. The calculations are so complex that a high-end PC would take a day to simulate a nanosecond (one-billionth of a second) worth of folding. At that rate, that PC would take 2,700 years to simulate a millisecond.
Blue Gene's advantage is tighter connections between the processors. The faster the processors can share information and distribute their workloads, the more complex research can be. IBM's Royyuru sees Blue Gene simulating not just the interaction of drugs and biological systems, but the interaction between molecules, the effect at the cellular level, in the tissue, the organs and so forth. Each step up reveals another layer of biological interconnection.
Finding a drug that matches a target the way a key matches a lock is known as "affinity," and it's something that leaps in computational power can help immediately. But speed is one thing; accuracy another. A key that fits a lock may also unlock all sorts of toxic reactions, or the door may swing open but not lead to a treatment any better than the previous one.
"That's why computing to a certain extent hasn't been such a breakthrough," says Ray Salemme, who has spent more than 30 years in labs, classrooms and boardrooms creating drugs with the help of the world's most sophisticated computers. "Everyone would like to close their eyes and have the answer pop out magically, but to get anything interesting you have to have intelligent people thinking."
Salemme contends that data storage and access is just as important as speed. With broadband and vast databases of research--such as the government's PubChem service--coming online, sophisticated processes are easier than ever.
In his Philadelphia-area home, Salemme has a hodgepodge of Apples and PCs with nearly a terabyte of disk storage. "As a hobby, I do protein engineering. I do things in my basement that would have required a major research lab 10 or 15 years ago," he says.
But compiling data and making sense of it are different propositions. Henry Markram is using a Blue Gene system to model the human brain, although his first task is making sense of years of data. "There is no model in brain research that can begin to integrate the smallest fraction of this data," said Markram, who runs the Laboratory of Neural Microcircuitry at the Ecole Polytechnique Federale de Lausanne, Switzerland.
With four Blue Gene racks, Markram aims to model the neurons of the neocortex column, the basic building block of the mammalian brain. Each neocortex column is half a millimeter in diameter and 1 millimeter to 3 millimeters high. A human neocortex column holds about 60,000 neurons, which look like trees whose roots and branches reach out and interconnect. That's 50 million to 100 million contact points per neocortex column.
The goal is to build a model that accurately represents the column and then make a model a viewer can actually walk through in three dimensions. There's a way to go, but Markram said that within a few years the project, called Blue Brain, could start to shed light on how autism works.
That would surely please IBM, not least because of the project's title.
fyi
fyi
science ping
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.