News News Supercomputer Boosts Scientists' Analytical Power of L-asparaginase. The complex structure of this enzyme, used to treat childhood leukemia, had eluded Wlodawer for 17 years. "Most of the progress has been made in the last 6 months," he said. "It resulted from a combination of better approaches, more determined work, and much better tools."
Gigabyte The problem used several weeks of computer time on the new Cray, and would have been far too complex and time-consuming for its predecessor, a Cray X-MP, Wlodawer said. The new Cray has a gigabyte of memory, or 128 million "words" compared with the older model's 4 million. The center, which is managed by PRI/Dyncorp, of Reston, Va., is housed at NCI's research center in Frederick,
Seeing Farther John Erickson, Ph.D., the center's scientific director, likened the supercomputer's value for biologists to that of a more powerful telescope for astronomers. "You can see faint objects, that you could barely make out before, much more clearly," Erickson said. "But more importantly, you can see farther out into the universe. You can ask questions you couldn't ask, and try experiments you couldn't really try before." Alexander Wlodawer, Ph.D., of Advanced BioScience Laboratories, Inc., an NCI contractor, said the computer recently allowed him and his coworkers to discern the crystal structure
Vol. 84, No. 6, March 18, 1992
Downloaded from http://jnci.oxfordjournals.org/ at Freie Universitaet Berlin on May 10, 2015
At the National Cancer Institute's Biomedical Supercomputing Center, state-of-the-art computer technology is helping to push the frontiers of biology. As biomedical scientists probe deeper into life's most fundamental mysteries, they depend increasingly on the kinds of high-tech tools, such as advanced computers, that have long been mainstays in physical science and engineering. Last September, the NCI center installed an eight-processor Cray Y-MP system, the most powerful computer currently available. The only supercomputer dedicated to biomedical research, the Cray is kept busy 24 hours a day, 7 days a week solving problems too big to be handled by standard computers. Users are about equally divided among NCI staff, other National Institutes of Health researchers, and scientists from around the world, who can communicate with the computer via networks such as INTERNET and NSFNET.
Md. Its facilities also include several smaller Vax and Convex computers, which serve in part as "front-end systems" through which users gain access to the Cray. Silicon Graphics workstations allow users to create graphic models of molecules and their interactions, and using specially designed goggles, to visualize these models in three-dimensional space. "What is emerging in all fields of science, including biology, is the ability to create a computational analogue of a natural system that applies first principles of science to produce simulations of phenomena not accessible by experiment," said Jacob Maizel, Ph.D., chief of NCI's Laboratory of Mathematical Biology. Maizel's observation is illustrated by Wlodawer's description of the supercomputer's role in refining the structure of L-asparaginase: "Once we have a rough model [from x-ray diffraction analysis], we try in the computer to 'heat' the molecule to a very high temperature, let everything move, and then hope that by cooling it down we can find the minimal energy conformation." Now that the enzyme's structure is known, Wlodawer said, researchers may be able to use that information to design new cancer drugs that share its therapeutic properties but not its side effects.
Computer-Intensive Future
Dr. Jacob Mabel
Inspired by the burgeoning field of nucleic acid sequence analysis, and encouraged by his own success in developing a computer technique for comparing molecules, Maizel brought his vision of a computer-intensive future for biomedical science when he came to NCI in 1983. The need for computers to handle the sheer quantities of raw sequence data being generated made converts of many scientists who were initially skeptical, Maizel said. "Biologists who had been quite refractory to the use of computers immediately
NEWS
385
News News
a surfeit of data, Maizel said. "Genome researchers are going to determine the sequence information in all these thousands of genes faster than we could conceivably be able to isolate and purify their products, and do crystallography and nuclear magnetic
386
"For a typical protein of 5,000 atoms, you have to calculate velocities for each atom at every incremental time point, and based on that you push the atom to a new position," Erickson explained. "This is an area that's computationally intense, partly because the time step that's re-
quired is so small — basically a femtosecond, or 10"15 second." Events such as two molecules binding, or the opening and closing of a loop, may occur on a time scale of microseconds, he said, which means that a billion time steps would have to be calculated in order to properly model the process. Determining the structures of molecules and the dynamics of their interactions is a crucial step in designing new drugs and optimizing the therapeutic properties of existing agents, Erickson said. His laboratory is currently studying the structure of an HIV protease inhibitor with clinical potential and is also working toward developing inhibitors for cathepsin D, an enzyme that is elevated in breast cancer patients and is suspected of being involved in metastasis.
New Machines As impressive as the Cray Y-MP's power and speed are, it is certain to be eclipsed in the not-too-distant future by newer machines that can perform even more prodigious feats.
Structure of the hydrophobic (brown) and hydrophiiic (yellow) interaction surfaces In a portion of the active site of HIV protease. Image generated by T. J. O'Donnell, O'Donncll & Associates.
Journal of the National Cancer Institute
Downloaded from http://jnci.oxfordjournals.org/ at Freie Universitaet Berlin on May 10, 2015
resonance," he said. "So we computational biologists feel there will be a major contribution in being able to read that sequence information and predict the properties encoded in it." Translating the genes into the corresponding amino acid sequences that make up their protein products is relatively straightAtomic structure of C2 symmetric inhibitor bound to the active site of HI V forward. But preprotease. The structure was determined by x-ray crystallography using a dicting the folding supercomputer. Image generated by Dr. John Erickson. of proteins — the three-dimensional sensed that they are necessary for hanstructure that is crucial in determining dling sequence data," he recalled. "Then, their functional properties — is far more having that data, we all want to know challenging. what it means." 'That's the holy grail of computaOne of the most powerful techniques tional biology," Maizel said. "People for answering this question involves cominitially thought there would be simple paring new data to a database of known rules like those governing Watson-Crick sequences, Maizel said. Computer base pairing in DNA, but so far those programs can check DNA and RNA serules have been very elusive." quences for identity or near identity, and "The computational challenge is astroperform sophisticated tests that reveal nomical, so that having a computer orders more subtle similarities. Newly disof magnitude more powerful than ever covered oncogenes, for example, are before makes a big impact," he added. analyzed in this way to discover relationMolecular dynamics introduces a ships to genes or proteins in humans, fourth dimension - time - to the study of animals, or viruses. chemical structures. Researchers apply basic laws of motion to create computer Accelerating Pace models of microscopic events too brief to be observed in the real world, such as an The accelerating pace of the Human enzyme reacting with a substrate or Genome Project in generating genetic ininhibitor. formation is giving molecular biologists
Hews
News
"Teraflop" Maizel said he is involved in a federal government project called the High Performance Computing Initiative, whose current goal is to demonstrate the feasibility of a "teraflop" computer. "What that means is that it can perform a million million floating point operations per second, which is two or three orders of magnitude more capability than we now have," he said. "Many areas of biology are stirring in anticipation at the thought of having that much computing power. Even with the problems and programs we have now, we could keep such a machine busy." Because gains in scientific knowledge inevitably generate ever more difficult questions, the demand for increasingly powerful tools like supercomputers is nearly insatiable, and computer designers are constantly pushing the limits of the machines' capabilities. "Our supercomputer has 1,000 million bytes of memory capacity, while neurobiologists say the human brain has the capacity to store a million million bytes of information," Maizel observed. "So we are far from reaching the point where a single computer simply stores as much data as a human, not to mention being able to convert that data into knowledge." —By Tom Reynolds
Vol. 84, No. 6, March 18, 1992
Search for Serum Tumor Markers Continues The discovery of alpha-fetoprotein (AFP) and carcinoembryonic antigen (CEA) in the 1960s sparked a search for other serum tumor markers that might indicate the presence of cancer in patients. Although low sensitivity and lack of specificity dashed initial hopes of using them as definitive tests for cancer, serum tumor markers have nevertheless become valuable instruments for monitoring results of therapy and recurrence of disease. In addition, research on the use of markers in combinations hints at improved monitoring and, perhaps, even more effective screens.
Recent Successes
not unique for cancer (see News, J Natl Cancer Inst,Feb. 19,1992). CEA is used in monitoring colon, lung, and breast cancer but can give false positive results in smokers or pregnant women. And some tumors, particularly the more aggressive, later-stage tumors, do not produce markers. "You can have negative results even in the presence of cancer," Taube said, "and you can have positive results in the absence of cancer." In addition, few of the useful markers are for the most important cancers. AFP, for example, is used for testicular and liver cancers, neither of which is in the top 10 U.S. cancers. And no effective marker exists for lung cancer, the number one cancer killer. Nevertheless, according to Jeffrey Schlom, Ph.D., chief of NCI's Laboratory of Tumor Immunology and Biology, investigators are learning to improve the
The mood is hopeful among investigators, bolstered by successes such as the discovery of organ-specific PSA (prostate-specific antigen), a marker for prostate cancer, in the late 1970s. But the mood is also tempered by frustrated expectations. "There is a slew of [markers] that we see coming into the literature with small feasibility tests that look very intriguing," said Sheila Taube, Ph.D., chief of NCI's Cancer Diagnosis Branch. "But they rarely pan out in larger studies." Many markers lack the specificity and sensitivity to be of great value. PS A is one of the most useful markers because it is produced by the prostate and nowhere else in the body, experts say. But because PSA can be increased by benign prostatic hypertrophy, the marker is Dr. Jeffrey Schiom
NEWS 387
Downloaded from http://jnci.oxfordjournals.org/ at Freie Universitaet Berlin on May 10, 2015
" 'Supercomputer' is a relative term," Erickson said. "Ten years from now, what we're calling supercomputers will probably be relics - things of the past." The supercomputing center staff must keep one eye on "technology that's coming down the road," he said. The hottest new field currently is parallel computing, which Erickson compared to having 10,000 brains working on a problem simultaneously. A conventional computer, in comparison, performs like a single brain that does calculations sequentially, albeit very fast.
Gigabyte The problem used several weeks of computer time on the new Cray, and would have been far too complex and time-consuming for its predecessor, a Cray X-MP, Wlodawer said. The new Cray has a gigabyte of memory, or 128 million "words" compared with the older model's 4 million. The center, which is managed by PRI/Dyncorp, of Reston, Va., is housed at NCI's research center in Frederick,
Seeing Farther John Erickson, Ph.D., the center's scientific director, likened the supercomputer's value for biologists to that of a more powerful telescope for astronomers. "You can see faint objects, that you could barely make out before, much more clearly," Erickson said. "But more importantly, you can see farther out into the universe. You can ask questions you couldn't ask, and try experiments you couldn't really try before." Alexander Wlodawer, Ph.D., of Advanced BioScience Laboratories, Inc., an NCI contractor, said the computer recently allowed him and his coworkers to discern the crystal structure
Vol. 84, No. 6, March 18, 1992
Downloaded from http://jnci.oxfordjournals.org/ at Freie Universitaet Berlin on May 10, 2015
At the National Cancer Institute's Biomedical Supercomputing Center, state-of-the-art computer technology is helping to push the frontiers of biology. As biomedical scientists probe deeper into life's most fundamental mysteries, they depend increasingly on the kinds of high-tech tools, such as advanced computers, that have long been mainstays in physical science and engineering. Last September, the NCI center installed an eight-processor Cray Y-MP system, the most powerful computer currently available. The only supercomputer dedicated to biomedical research, the Cray is kept busy 24 hours a day, 7 days a week solving problems too big to be handled by standard computers. Users are about equally divided among NCI staff, other National Institutes of Health researchers, and scientists from around the world, who can communicate with the computer via networks such as INTERNET and NSFNET.
Md. Its facilities also include several smaller Vax and Convex computers, which serve in part as "front-end systems" through which users gain access to the Cray. Silicon Graphics workstations allow users to create graphic models of molecules and their interactions, and using specially designed goggles, to visualize these models in three-dimensional space. "What is emerging in all fields of science, including biology, is the ability to create a computational analogue of a natural system that applies first principles of science to produce simulations of phenomena not accessible by experiment," said Jacob Maizel, Ph.D., chief of NCI's Laboratory of Mathematical Biology. Maizel's observation is illustrated by Wlodawer's description of the supercomputer's role in refining the structure of L-asparaginase: "Once we have a rough model [from x-ray diffraction analysis], we try in the computer to 'heat' the molecule to a very high temperature, let everything move, and then hope that by cooling it down we can find the minimal energy conformation." Now that the enzyme's structure is known, Wlodawer said, researchers may be able to use that information to design new cancer drugs that share its therapeutic properties but not its side effects.
Computer-Intensive Future
Dr. Jacob Mabel
Inspired by the burgeoning field of nucleic acid sequence analysis, and encouraged by his own success in developing a computer technique for comparing molecules, Maizel brought his vision of a computer-intensive future for biomedical science when he came to NCI in 1983. The need for computers to handle the sheer quantities of raw sequence data being generated made converts of many scientists who were initially skeptical, Maizel said. "Biologists who had been quite refractory to the use of computers immediately
NEWS
385
News News
a surfeit of data, Maizel said. "Genome researchers are going to determine the sequence information in all these thousands of genes faster than we could conceivably be able to isolate and purify their products, and do crystallography and nuclear magnetic
386
"For a typical protein of 5,000 atoms, you have to calculate velocities for each atom at every incremental time point, and based on that you push the atom to a new position," Erickson explained. "This is an area that's computationally intense, partly because the time step that's re-
quired is so small — basically a femtosecond, or 10"15 second." Events such as two molecules binding, or the opening and closing of a loop, may occur on a time scale of microseconds, he said, which means that a billion time steps would have to be calculated in order to properly model the process. Determining the structures of molecules and the dynamics of their interactions is a crucial step in designing new drugs and optimizing the therapeutic properties of existing agents, Erickson said. His laboratory is currently studying the structure of an HIV protease inhibitor with clinical potential and is also working toward developing inhibitors for cathepsin D, an enzyme that is elevated in breast cancer patients and is suspected of being involved in metastasis.
New Machines As impressive as the Cray Y-MP's power and speed are, it is certain to be eclipsed in the not-too-distant future by newer machines that can perform even more prodigious feats.
Structure of the hydrophobic (brown) and hydrophiiic (yellow) interaction surfaces In a portion of the active site of HIV protease. Image generated by T. J. O'Donnell, O'Donncll & Associates.
Journal of the National Cancer Institute
Downloaded from http://jnci.oxfordjournals.org/ at Freie Universitaet Berlin on May 10, 2015
resonance," he said. "So we computational biologists feel there will be a major contribution in being able to read that sequence information and predict the properties encoded in it." Translating the genes into the corresponding amino acid sequences that make up their protein products is relatively straightAtomic structure of C2 symmetric inhibitor bound to the active site of HI V forward. But preprotease. The structure was determined by x-ray crystallography using a dicting the folding supercomputer. Image generated by Dr. John Erickson. of proteins — the three-dimensional sensed that they are necessary for hanstructure that is crucial in determining dling sequence data," he recalled. "Then, their functional properties — is far more having that data, we all want to know challenging. what it means." 'That's the holy grail of computaOne of the most powerful techniques tional biology," Maizel said. "People for answering this question involves cominitially thought there would be simple paring new data to a database of known rules like those governing Watson-Crick sequences, Maizel said. Computer base pairing in DNA, but so far those programs can check DNA and RNA serules have been very elusive." quences for identity or near identity, and "The computational challenge is astroperform sophisticated tests that reveal nomical, so that having a computer orders more subtle similarities. Newly disof magnitude more powerful than ever covered oncogenes, for example, are before makes a big impact," he added. analyzed in this way to discover relationMolecular dynamics introduces a ships to genes or proteins in humans, fourth dimension - time - to the study of animals, or viruses. chemical structures. Researchers apply basic laws of motion to create computer Accelerating Pace models of microscopic events too brief to be observed in the real world, such as an The accelerating pace of the Human enzyme reacting with a substrate or Genome Project in generating genetic ininhibitor. formation is giving molecular biologists
Hews
News
"Teraflop" Maizel said he is involved in a federal government project called the High Performance Computing Initiative, whose current goal is to demonstrate the feasibility of a "teraflop" computer. "What that means is that it can perform a million million floating point operations per second, which is two or three orders of magnitude more capability than we now have," he said. "Many areas of biology are stirring in anticipation at the thought of having that much computing power. Even with the problems and programs we have now, we could keep such a machine busy." Because gains in scientific knowledge inevitably generate ever more difficult questions, the demand for increasingly powerful tools like supercomputers is nearly insatiable, and computer designers are constantly pushing the limits of the machines' capabilities. "Our supercomputer has 1,000 million bytes of memory capacity, while neurobiologists say the human brain has the capacity to store a million million bytes of information," Maizel observed. "So we are far from reaching the point where a single computer simply stores as much data as a human, not to mention being able to convert that data into knowledge." —By Tom Reynolds
Vol. 84, No. 6, March 18, 1992
Search for Serum Tumor Markers Continues The discovery of alpha-fetoprotein (AFP) and carcinoembryonic antigen (CEA) in the 1960s sparked a search for other serum tumor markers that might indicate the presence of cancer in patients. Although low sensitivity and lack of specificity dashed initial hopes of using them as definitive tests for cancer, serum tumor markers have nevertheless become valuable instruments for monitoring results of therapy and recurrence of disease. In addition, research on the use of markers in combinations hints at improved monitoring and, perhaps, even more effective screens.
Recent Successes
not unique for cancer (see News, J Natl Cancer Inst,Feb. 19,1992). CEA is used in monitoring colon, lung, and breast cancer but can give false positive results in smokers or pregnant women. And some tumors, particularly the more aggressive, later-stage tumors, do not produce markers. "You can have negative results even in the presence of cancer," Taube said, "and you can have positive results in the absence of cancer." In addition, few of the useful markers are for the most important cancers. AFP, for example, is used for testicular and liver cancers, neither of which is in the top 10 U.S. cancers. And no effective marker exists for lung cancer, the number one cancer killer. Nevertheless, according to Jeffrey Schlom, Ph.D., chief of NCI's Laboratory of Tumor Immunology and Biology, investigators are learning to improve the
The mood is hopeful among investigators, bolstered by successes such as the discovery of organ-specific PSA (prostate-specific antigen), a marker for prostate cancer, in the late 1970s. But the mood is also tempered by frustrated expectations. "There is a slew of [markers] that we see coming into the literature with small feasibility tests that look very intriguing," said Sheila Taube, Ph.D., chief of NCI's Cancer Diagnosis Branch. "But they rarely pan out in larger studies." Many markers lack the specificity and sensitivity to be of great value. PS A is one of the most useful markers because it is produced by the prostate and nowhere else in the body, experts say. But because PSA can be increased by benign prostatic hypertrophy, the marker is Dr. Jeffrey Schiom
NEWS 387
Downloaded from http://jnci.oxfordjournals.org/ at Freie Universitaet Berlin on May 10, 2015
" 'Supercomputer' is a relative term," Erickson said. "Ten years from now, what we're calling supercomputers will probably be relics - things of the past." The supercomputing center staff must keep one eye on "technology that's coming down the road," he said. The hottest new field currently is parallel computing, which Erickson compared to having 10,000 brains working on a problem simultaneously. A conventional computer, in comparison, performs like a single brain that does calculations sequentially, albeit very fast.
