Trade publication covering breaking news in the rapidly evolving supercomputer marketplace.
  1. Researchers at Linköping University in Sweden have developed a method to increase by a factor of five the computing power of a standard algorithm when performed in one type of standard chip. The new method is both simple and smart, but the road to publication has been long.  

    We are dealing with a programmable integrated circuit known as an "FPGA", which is an abbreviation for "field-programmable gate array". This consists of a matrix of logical gates that can be programmed in situ, and can be reprogrammed an unlimited number of times. The first FPGAs came onto the market in 1985, and sales since then have increased dramatically. The market is now dominated by a couple of major players, and is expected to amount to USD 9.8 billion in 2020 (Wikipedia). The researchers have increased in these chips the speed of an algorithm known as the "fast Fourier transform", which is used in spectral analysis, radar technology and telecommunication.

    "Until now, people have believed that once an FPGA is full it cannot accommodate any more. If you want new functionality in this case, you have to completely rebuild the hardware, which is expensive," says Oscar Gustafsson, senior lecturer in the Department of Computer Engineering at Linköping University.

    But Carl Ingemarsson, a PhD student at the department, had other ideas. As an undergraduate several years ago, he was challenged to increase the speed of calculation in an FPGA. If the lab group could manage to reach a frequency greater than 450 MHz, they wouldn't have to carry out the final lab in the course.

    "This was what was needed to convince me to examine in depth the way the logic is represented inside the chip," he says.

    He achieved the frequency, skipped the final lab, and at the same time laid the foundation for his doctoral project. The result is that FPGAs today can be made to work five times as fast, or to deal with five times the number of calculations. While it's true that Carl has only confirmed this in two families of FPGA, there is no reason to believe that it is not also the case for all other families.

    "This advance will save huge sums for demanding calculations in industry, and will make it possible to implement new functionality without needing to replace the hardware," says Oscar Gustafsson.

    Carl Ingemarsson's method is based on ensuring that the signal takes a smarter route through the various building blocks inside the chip.

    "Normally, you choose an algorithm that can carry out the desired calculations, and then build up the structure, the architecture, using the required blocks. This is then transferred to the FPGA. But we have also looked at how the logic is built up, the routes the signals take, and what happens to them inside the chip. We have then adapted the architecture and the mapping onto the chip using the results of this analysis."

    A clever change in the signal routes gives the chip a capacity that is five times greater for each hardware unit.

    "It should be possible to automate this optimisation of the chip," says Carl Ingemarsson.

    The method was, however too simple, or too ingenious, for the scientific reviewers.

    "At one level, it might seem that we haven't changed anything, we're still using the same standard components, but we have increased the computing power by a factor of five. This has made it has difficult to get our article published in a scientific journal," Oscar Gustafsson explains.

    But the solution was so clever that someone managed to plagiarise the work before the IEEE decided to publish it. It suddenly appeared at an IEEE conference, using copies of the diagrams, with parts of the text swapped out and completely different authors. All the support documentation in the form of original files and original diagrams was, however, available at LiU: the plagiarism was discovered, and the researcher suspended. The damage had been done, however, and publication of the original article was delayed by at least a year.

    In the meantime, Carl Ingemarsson while waiting to be able to complete his doctoral thesis has started to work at Ericsson. An academic career doesn’t appear to be so attractive any longer.

    “My wife and I are planning on starting a microbrewery, so that when the thesis is finally presented I will be able to offer beer I have brewed myself,” he promises.

  2. IMG 3038 088cd

    Proteins are the building blocks of life and biological agents. They are drivers of growth and development and the spread of viruses and bacteria, and have key roles in disease pathways and virtually all cellular functions. As scientists gain knowledge about proteins, the mechanisms behind biological mysteries are revealed. 

    To help shed light on the workings of proteins, Virginia Commonwealth University researcher Lukasz Kurgan, Ph.D., vice chair of the Computer Science Department in the School of Engineering, has developed a series of bioinformatics programs to assist biologists in developing insights into the functions of intrinsically disordered proteins. This group of proteins lacks a fixed structure, which means they are totally or partially flexible and amorphous.

    Over the last several decades, scientists have sequenced 85 million unique proteins, structured and unstructured alike, but still don’t know what the vast majority of these proteins do. As more proteins are discovered, more sophisticated supercomputer programs must be developed to help determine their functions.

    “We have manually curated but understand less than 1 percent of these proteins, and right now there’s over 80 million to solve,” said Kurgan, a Qimonda-endowed professor and data scientist. “A program can solve these proteins faster than a single human and can help researchers speed up hypothesis generation.”

    Solving the puzzle 

    Determining a protein’s function becomes even more challenging when a protein is completely or partially disordered. When a protein does have a defined structure, researchers use prior knowledge and bioinformatics programs to first decipher that structure, which then helps determine function. If the protein is disordered, biologists turn to programs built by Kurgan and other computer scientists that use predictive models to generate workable hypotheses on the protein’s function.

    Since 2008, Kurgan has developed four programs for this purpose. This spring, his team was awarded a $500,000 grant from the National Science Foundation to develop subsequent programs. So far, Kurgan’s programs have more than 7,000 users from more than 1,300 cities in 96 countries.

    Kurgan has also developed six programs that determine whether a protein is disordered or not. In 2012, his MFDp program was ranked third out of 28 participants in the biannual worldwide CASP10 experiment, which evaluates the effectiveness of computer and human predictors of intrinsic disorder. In 2014, Kurgan’s lab released DisoRDPbind, the first program to predict multiple functions of intrinsically disordered proteins.

    Kurgan’s programs use existing collections of data on proteins whose functions have been determined to build predictive models to map the functions of unknown intrinsically disordered proteins.

    “The details are not easy. Building these models takes a little bit of art, theory and experience,” Kurgan said.

    New ideas for disorder

    It is a commonly accepted fact among scientists that disordered proteins, similar to their structured counterparts, have essential functions. This assertion was at first met with disbelief, as is initially common with many scientific discoveries.

    “About 30 years ago, when disordered proteins were discovered, there were a lot of deniers. Some people said that this is just noise in the protein structures,” Kurgan said. “Now, disorder as a mechanism of biology is an accepted fact. Just because a protein has no defined structure, doesn’t mean it’s useless. It just works in a different way.”

    Now, deciphering disorder is a collaborative effort in the scientific community, and various programs from multiple entities come together to provide different approaches to determine the functions of disordered proteins. Kurgan has worked with researchers from the University of South Florida, Indiana University, and Tianjin and Nankai Universities in China, on a study that used his programs to discover the incidence of intrinsic disorder in close to 1,000 species from all kingdoms of life. Several other collaborative studies have focused on the functional roles of intrinsic disorder in HIV, Hepatitis C and Dengue viruses.

    “Collectively we can push the boundaries of what is being done,” Kurgan said. “It’s not based on the efforts of one specific researcher or group. Collectively we help each other.”

  3. It was midafternoon, but it was dark in an area in Boulder, Colorado on Aug. 3, 1998. A thick cloud appeared overhead and dimmed the land below for more than 30 minutes. Well-calibrated radiometers showed that there were very low levels of light reaching the ground, sufficiently low that researchers decided to simulate this interesting event with supercomputer models. Now in 2017, inspired by the event in Boulder, NASA scientists will explore the moon's eclipse of the sun to learn more about Earth's energy system. 

    {media id=170,layout=blog}

    On Aug. 21, 2017, scientists are looking to this year's total solar eclipse passing across America to improve our modelling capabilities of Earth's energy. Guoyong Wen, a NASA scientist working for Morgan State University in Baltimore, is leading a team to gather data from the ground and satellites before, during and after the eclipse so they can simulate this year's eclipse using an advanced supercomputer model, called a 3-D radiative transfer model. If successful, Wen and his team will help develop new calculations that improve our estimates of the amount of solar energy reaching the ground, and our understanding of one of the key players in regulating Earth's energy system, clouds.

    Earth's energy system is in a constant dance to maintain a balance between incoming radiation from the sun and outgoing radiation from Earth to space, which scientists call the Earth's energy budget. The role of clouds, both thick and thin, is important in their effect on energy balance.

    Like a giant cloud, the moon during the 2017 total solar eclipse will cast a large shadow across a swath of the United States. Wen and his team already know the dimensions and light-blocking properties of the moon, but will use ground and space instruments to learn how this large shadow affects the amount of sunlight reaching Earth's surface, especially around the edges of the shadow.

    "This is the first time we're able to use measurements from the ground and from space to simulate the moon's shadow going across the face of Earth in the United States and calculating energy reaching the Earth," said Wen. Scientists have made extensive atmospheric measurements during eclipses before, but this is the first opportunity to collect coordinated data from the ground and from a spacecraft that observes the entire sunlit Earth during an eclipse, thanks to the National Oceanic and Atmospheric Administration's Deep Space Climate Observatory launched (DSCOVR) in February 2015.

    Even though the moon blocking the sun during a solar eclipse and clouds blocking sunlight to Earth's surface are two different phenomena, both require similar mathematical calculations to accurately understand their effects. Wen anticipates this experiment will help improve the current model calculations and our knowledge of clouds, specifically thicker, low altitude clouds that may cover about 30 percent of the planet at any given time.

    In this experiment, Wen and his team will simulate the total solar eclipse in a 3-D radiative transfer model, which helps scientists understand how energy is propagated on Earth. Currently, models tend to depict clouds in one dimension. In many cases, these one dimensional calculations can create useful science models for understanding the atmosphere. Sometimes though, a three-dimensional calculation is needed to provide more accurate results. The big difference is that 3-D clouds reflect or scatter solar energy in many directions, from the top and bottom, and also out of the sides of clouds. This 3-D behavior results in different amounts of energy reaching the ground than a one-dimensional model could predict.

    "We're testing the ability to do a certain kind of complex calculation, a test of a 3-D mathematical technique, to see if this is an improvement over the previous technique," said Jay Herman, scientist at NASA's Goddard Space Flight Center in Greenbelt, Maryland, and co-investigator of the project. "If this is successful, then we will have a better tool to implement in climate models and can use it to answer questions and the Earth's energy budget and climate." For the upcoming eclipse, Wen and his team members will be stationed in Casper, Wyoming, and Columbia, Missouri to gather information on the amount of energy being transmitted to and from Earth before, during and right after the eclipse with several ground instruments.

    A ground-based, NASA-developed Pandora Spectrometer Instrument will provide information on how much of any given wavelength of light is present, and a pyranometer will measure total solar energy from all directions coming down toward the surface. Immediately before and after the eclipse scientists will measure other information such as the amount of absorbing trace gases in the atmosphere, such as ozone, nitrogen dioxide and small aerosol particles to also use in the 3-D model.

    Meanwhile in space, NASA's Earth Polychromatic Imaging Camera, or EPIC, instrument aboard the DSCOVR spacecraft, will observe the light leaving Earth and allow scientists to estimate of the amount of light reaching the earth's surface. Additionally, NASA's two MODIS satellite instruments, aboard the agency's Terra and Aqua satellites, launched in 1999 and 2002, respectively, will provide observations of atmospheric and surface conditions at times before and after the eclipse. The scientists will then combine ground measurements with those observed by the spacecraft.

    This experiment complements NASA's decades-long commitment to observing and understanding contributions to Earth's energy budget. For more than 30 years, NASA has measured and calculated the amount of solar energy hitting the top of our atmosphere, the amount of the sun's energy reflected back to space and how much thermal energy is emitted by our planet to space. These measurements have been possible thanks to instruments and missions such as ACRIMSAT and SOLSTICE (launched in 1991), and SORCE, launched in 2003 as well as the series of CERES instruments flown aboard Terra, Aqua, and Suomi-NPP (launched in 2011).

    This fall, NASA will continue to monitor the sun-Earth relationship by launching the Total and Spectral Solar Irradiance Sensor-1, or TSIS-1, to the International Space Station and the sixth Clouds and the Earth's Radiant Energy System CERES instrument, CERES FM6, to orbit later this year. Five CERES instruments are currently on orbit aboard three satellites.

  4. Near real-time monitoring could enable faster response to outbreaks

    An analytical tool that combines Google search data with government-provided clinical data can quickly and accurately track dengue fever in less-developed countries, according to new research published in PLOS Computational Biology by Shihao Yang of Harvard University and colleagues.

    The research builds on a methodology previously developed by the team to track influenza in the United States. This mathematical modeling tool, known as "AutoRegression with GOogle search queries" (ARGO), revived hopes in 2015 that internet search data could help health officials track diseases after earlier systems like Google Flu Trends and Google Dengue Trends returned poor results.

    In the new study, the research team modified ARGO to explore its potential to track dengue activity in Mexico, Brazil, Thailand, Singapore, and Taiwan. Dengue, a mosquito-borne virus that infects about 390 million people each year, is often difficult to monitor with traditional hospital-based reporting due to inefficient communication, but dengue-related Google searches could provide faster alerts.

    The researchers used Google's "Trends" tool to track the top ten dengue-related search queries made by users in each country during the study period. They also gathered historical dengue data from government health agencies and input both datasets into ARGO. Using the assumption that more dengue-related searches occur when more people are infected, ARGO calculated near real-time estimates of dengue prevalence for each country.

    The scientists then compared ARGO's estimates with those from five other methods. They found that ARGO returned more accurate estimates than did any other method for Mexico, Brazil, Thailand, and Singapore. Estimates for Taiwan were less accurate, possibly because the country experienced less-consistent seasonal disease patterns from year to year.

    The findings highlight the potential for Google searches to enable accurate, timely tracking of mosquito-borne diseases in countries lacking effective traditional surveillance systems. Future work could investigate whether this method could be improved to track disease on finer spatial and temporal scales, and whether environmental data, such as temperature, could improve estimates.

    "The wide availability of internet throughout the globe provides the potential for an alternative way to reliably track infectious diseases, such as dengue, faster than traditional clinical-based systems," says study senior author Mauricio Santillana of Boston Children's Hospital and Harvard Medical School. "This alternative way of tracking disease could be used to alert governments and hospitals when elevated dengue incidence is anticipated, and provide safety information for travelers."

  5. Findings have implications for quantum supercomputing

    Researchers have performed the first ever quantum-mechanical simulation of the benchmark ultracold chemical reaction between potassium-rubidium (KRb) and a potassium atom, opening the door to new controlled chemistry experiments and quantum control of chemical reactions that could spark advances in quantum supercomputing and sensing technologies. The research by a multi-institutional team simulated the ultracold chemical reaction, with results that had not been revealed in experiments. 

    "We found that the overall reactivity is largely insensitive to the underlying chaotic dynamics of the system," said Brian Kendrick of Los Alamos National Laboratory's Theoretical Division, "This observation has important implications for the development of controlled chemistry and for the technological applications of ultracold molecules from precision measurement to quantum computing."

    The research addressed open questions about whether chemical reactions occur at a billionth of a degree above absolute zero and whether the outcome can be controlled. Scientists worldwide are addressing these questions experimentally by cooling and trapping atoms and molecules at temperatures close to absolute zero and allowing them to interact chemically. This field of chemistry, widely referred to as ultracold chemistry, has become a hotbed for controlled chemistry experiments and quantum control of chemical reactions, the holy grail of chemistry. 

    In a pioneering experiment in 2010, groups at Colorado's JILA (formerly known as the Joint Institute for Laboratory Astrophysics) were able to produce an ultracold gas of KRb molecules at nano-Kelvin temperatures. By merely flipping the nuclear spin of a KRb molecule they demonstrated that the ultracold reaction between these molecules could be turned on or off -- a perfect illustration of controlled on-demand chemistry.

    But theoretical calculations of the reaction dynamics for such systems pose a daunting computational challenge. The calculations of the K + KRb reaction provide new insights into the reaction dynamics that is not revealed in the experiments -- that the rotationally resolved reaction rates exhibit a statistical (Poisson) distribution. 

    A fascinating finding of their study, Kendrick notes, is that while the overall reactivity is governed by the long-range forces, the rotational populations of the product K2 molecule are governed by chaotic dynamics at short-range. "The chaotic dynamics appears to be a general property of all ultracold reactions involving heavy alkali molecules," said Kendrick, "so the rotational populations of all such reactions will exhibit the same Poisson distribution."

    This new, fundamental understanding of ultracold reactions will guide related technological applications in quantum control/computing, precision measurement and sensing important to the Los Alamos missions in information science and technology and global security.

  6. Scientists have trained a computer to recognize beautiful scenery using “deep learning”, an approach to artificial intelligence which is inspired by the architecture of the human brain. 

    Researchers at the Data Science Lab at Warwick Business School took more than 200,000 images of places in the UK that had been rated for their beauty on the website Scenic-or-Not and showed them to a deep learning model in order to find out what makes a scenic location beautiful. 

    The deep learning model processed all 200,000 images and labelled them with information on what was in the picture, such as “valley”, “grass”, “no horizon” or “open space”. Using these labels, the researchers were able to investigate which attributes of a scene led to higher scenic scores. 

    The scientists then trained a new deep learning model to look at pictures and rate them itself. 

    Chanuki Seresinhe, of the Data Science Lab at Warwick Business School, said: “We tested our model in London and it not only identified parks like Hampstead Heath as beautiful, but also built-up areas such as Big Ben and the Tower of London.” 

    Ms Seresinhe, along with WBS Data Science Lab directors Suzy Moat and Tobias Preis, used the MIT Places Convolutional Neural Network – a deep learning model – to analyse the images from Scenic-or-Not, which were rated by 1.5 million people, and find what attributes, such as “trees”, “mountain”, “hospital” and “highway”, corresponded to high and low scenic ratings. 

    Deep learning models are a particular kind of “neural network” – simulated networks of neurons, like those in the human brain – and have driven recent dramatic advances in artificial intelligence tasks, such as facial recognition and speech recognition. 

    Using the MIT Places deep learning model, the researchers found that features such as "valley", "coast", "mountain" and "trees" were associated with higher scenicness. 

    However, some man-made elements also tended to improve scores, including historical architecture such as "church", "castle", "tower" and "cottage", as well as bridge-like structures such as "viaduct" and "aqueduct". Interestingly, large areas of greenspace such as "grass" and "athletic field" led to lower ratings of scenicness rather than boosting scores. 

    Ms Seresinhe added: “It appears that the old adage ‘natural is beautiful’ seems to be incomplete: flat and uninteresting green spaces are not necessarily beautiful, while characterful buildings and stunning architectural features can improve the beauty of a scene.

    “I am fascinated by how deep learning can help us to develop a deeper insight into what human beings collectively might understand to be beautiful.” 

    Dr Moat, Associate Professor of Behavioural Science at Warwick Business School and co-director of the Data Science Lab, said: ”These findings are of particular interest in the context of our previous research, which showed that people who live in areas rated as more scenic report their health to be better, even when we take data on greenspace into account. 

    “Our new results shine light on why a location being green might not be enough for it to be considered attractive. This distinction has clear relevance for planning decisions which aim to improve the wellbeing of local inhabitants.”

    The scientists then adapted the deep learning model to rate the scenicness of new locations, and tested it on more than 200,000 photographs of London that the model hadn’t seen before. 

    Professor Preis, Professor of Behavioural Science and Finance at Warwick Business School and co-director of the Data Science Lab, said: “It was fascinating to see that the model understood that bridges and historical architecture increase the perceived beauty of a scene, while grass and greenery is not necessarily scenic. 

    “Our previous results make it clear that scientists and policymakers alike need measurements of environmental beauty, not just measurements of how green places are. Games like Scenic-or-Notcan help us collect millions of ratings from humans, but having a model which can automatically tell us whether a place is beautiful or not opens up completely new horizons.” 

    The paper, Using deep learning to quantify the beauty of outdoor places, by Ms Seresinhe, Professor Preis and Professor Moat is published in Royal Society Open Science.

  7. Revenue of $19.3 billion, down for the 21st quarter in a row, off by more than 4 percent

    IBM stock was down 2.5 percent after the company announced earnings for the second quarter of the year, falling short of revenue estimates but beating earnings estimates. Consolidated diluted earnings per share were $4.32 compared to $4.69, down 8 percent year to year. Consolidated net income was $4.1 billion compared to $4.5 billion in the year-ago period, a decrease of 10 percent. Revenues from continuing operations for the six-month period totaled $37.4 billion, a decrease of 4 percent year to year (decrease of 3 percent adjusting for currency) compared with $38.9 billion for the first six months of 2016.

    Operating (non-GAAP) diluted earnings per share from continuing operations were $5.35 compared with $5.30 per diluted share for the 2016 period, an increase of 1 percent. Operating (non-GAAP) net income for the six months ended June 30, 2017 was $5.0 billion compared with $5.1 billion in the year-ago period, a decrease of 1 percent.

    • EPS: Excluding certain items, $2.97 in earnings per share vs. $2.74 in earnings per share as expected by analysts, according to Thomson Reuters.
    • Revenue: $19.29 billion vs. $19.46 billion as expected by analysts, according to Thomson Reuters.
    • Segment Results for Second Quarter

      • Cognitive Solutions (includes solutions software and transaction processing software) -- revenues of $4.6 billion, down 2.5 percent (down 1.4 percent adjusting for currency). Pre-tax income increased at a double-digit rate.
      • Global Business Services (includes consulting, global process services and application management) --revenues of $4.1 billion, down 3.7 percent (down 1.7 percent adjusting for currency). Strategic imperatives grew 8 percent led by the cloud and mobile practices.
      • Technology Services & Cloud Platforms (includes infrastructure services, technical support services and integration software) -- revenues of $8.4 billion, down 5.1 percent (down 3.6 percent adjusting for currency). Strategic imperatives, driven by hybrid cloud services, grew 20 percent.
      • Systems (includes systems hardware and operating systems software) -- revenues of $1.7 billion, down 10.4 percent (down 9.6 percent adjusting for currency).
      • Global Financing (includes financing and used equipment sales) -- revenues of $415 million, down 2.2 percent (down 1.7 percent adjusting for currency).

    "In the second quarter, we strengthened our position as the enterprise cloud leader and added more of the world's leading companies to the IBM Cloud," said Ginni Rometty, IBM chairman, president and chief executive officer. "We continue to innovate, adding regtech capabilities to our portfolio of Watson offerings; developing solutions based on emerging technologies such as Blockchain; and reinventing the IBM mainframe by enabling clients to encrypt all data, all the time.”

          SECOND QUARTER 2017
                  Gross Profit    
          Diluted EPS   Net Income   Margin    
      GAAP from Continuing Operations   $2.48   $2.3B   45.6%    
      Year/Year   -5%   -7%   -2.3Pts    
      Operating (Non-GAAP)   $2.97   $2.8B   47.2%    
      Year/Year   1%   -2%   -1.8Pts    
              Strategic       annual exit
      REVENUE   Total IBM   Imperatives   Cloud   run rate
      As reported (US$)   $19.3B   $8.8B   $3.9B   $8.8B
      Year/Year   -5%   5%   15%   30%
      Year/Year adjusting for currency   -3%   7%   17%   32%

    "We finished the first half of the year where we expected, including continued strong free cash flow generation," said Martin Schroeter, IBM senior vice president and chief financial officer. "This allowed us to continue our strong R&D investment levels and return more than $5 billion to shareholders through dividends and gross share repurchases during the first half."

    Strategic Imperatives

    Second-quarter cloud revenues increased 15 percent (up 17 percent adjusting for currency) to $3.9 billion. Cloud revenue over the last 12 months was $15.1 billion. The annual exit run rate for as-a-service revenue increased to $8.8 billion from $6.7 billion in the second quarter of 2016. Revenues from analytics increased 4 percent (up 6 percent adjusting for currency). Revenues from mobile increased 27 percent (up 29 percent adjusting for currency) and revenues from security increased 4 percent (up 5 percent adjusting for currency).

    Full-Year 2017 Expectations

    The company continues to expect operating (non-GAAP) diluted earnings per share of at least $13.80 and GAAP diluted earnings per share of at least $11.95. Operating (non-GAAP) diluted earnings per share exclude $1.85 per share of charges for amortization of purchased intangible assets, other acquisition-related charges and retirement-related charges. IBM continues to expect free cash flow to be relatively flat year to year.

    Cash Flow and Balance Sheet

    In the second quarter, the company generated net cash from operating activities of $3.5 billion; or $3.3 billion excluding Global Financing receivables. IBM’s free cash flow was $2.6 billion. IBM returned $1.4 billion in dividends and $1.4 billion of gross share repurchases to shareholders. At the end of June 2017, IBM had $2.4 billion remaining in the current share repurchase authorization.

    IBM ended the second quarter of 2017 with $12.3 billion of cash on hand. Debt totaled $45.7 billion, including Global Financing debt of $29.0 billion. The balance sheet remains strong and, with the completion of the reorganization of its client and commercial financing business, the company is better positioned over the long term.

  8. A smart supercomputer program named JAABA has helped scientists create a brain-wide atlas of fruit fly behavior.

    The machine-learning program tracked the position and cataloged the behaviors of 400,000 fruit flies, in more than 225 days of video footage, helping researchers match specific behaviors to different groups of neurons. 

    {media id=169,layout=blog}

    "We wanted to understand what neurons are doing at the cellular level," says Janelia Group Leader Kristin Branson. She and colleagues reported the work July 13 in the journal Cell.

    Their results are the most comprehensive neural maps of behavior yet created. Such detailed maps could give researchers a starting point for tracing the neural circuitry flies use to produce specific behaviors, such as jumping or wing grooming, Branson says. Understanding the inner workings of the fly brain could even offer insight into the neural basis of human behavior, she says.

    Though the brain of the fruit fly, Drosophila melanogaster, is only about the size of a poppy seed, it comprises roughly 100,000 neurons which interact in complex circuits to control an extensive array of behaviors.

    "Flies do all the things that an organism needs to do in the world," says study coauthor Alice Robie, a research scientist at Janelia. "They have to find food, they have to escape from predators, they have to find a mate, they have to reproduce." All those actions, she says, involve different behaviors for interacting with the environment.

    Scientists have identified some of the neurons at work in courtship, say, or chasing, but no one has tackled the entire brain all at once. Branson's team took a brain-wide approach for finding neurons involved in a suite of 14 behaviors, including wing flicking, crab walking, and attempted copulation.

    The team studied 2,204 populations of flies, part of a collection developed at Janelia called the GAL4 Fly Lines. The flies are genetically engineered to crank up the activity of certain neurons. Previous imaging work, Janelia's FlyLight Project, identified where in the brain these neurons resided - so researchers already had an anatomical map of the neurons targeted in each group of flies. But researchers didn't know what role these neurons played in behavior.

    Dialing up the neurons' activity in one type of flies, for example, made them huddle together when placed in a shallow dish, says lab technician Jonathan Hirokawa, now a mechatronics engineer at Rockefeller University in New York City. Other types of flies acted even more bizarrely, he recalls. "Sometimes you'd get flies that would all turn in circles, or all follow one another like they were in a conga line."

    From these behavioral quirks, researchers could piece together the cell types involved in walking or backing up, for example. The researchers tackled the problem in an automated fashion, Robie says. Using videos of flies, Robie taught the machine-vision and -learning program JAABA, Janelia Automatic Animal Behavior Annotator, how to recognize specific behaviors. Then Branson's team put JAABA to work watching and labeling behaviors in videos of the 2,204 different fly groups - a feat that would have taken humans some 3,800 years.

    In addition to matching cell types to behaviors, the researchers identified something entirely new: the nerve cells linked to female chase behavior. "There have been some reports of female aggression, but not females chasing other flies," Robie says.

    That finding stands out, Branson says, but it's just one of thousands of results yielded by their study. "With these big datasets, we've been trying to figure out how you actually share the information," she says. Their solution is a program called BABAM, or the Browsable Atlas of Behavior Anatomy Maps. With BABAM, scientists can explore the new data, create maps that link behavior to fly brain anatomy, and search for fly groups associated with certain behaviors.

    Branson and Robie say the new results highlight the advantages of blending different scientific disciplines at Janelia. "This is what happens when you put biologists and computer scientists together," Robie says.

  9. A research team led by scientists at UC San Francisco has developed a supercomputational method to systematically probe massive amounts of open-access data to discover new ways to use drugs, including some that have already been approved for other uses.

    The method enables scientists to bypass the usual experiments in biological specimens and to instead do computational analyses, using open-access data to match FDA-approved drugs and other existing compounds to the molecular fingerprints of diseases like cancer. The specificity of the links between these drugs and the diseases they are predicted to be able to treat holds the potential to target drugs in ways that minimize side effects, overcome resistance and reveal more clearly how both the drugs and the diseases are working.

    "This points toward a day when doctors may treat their patients with drugs that have been individually tailored to the idiosyncracies of their own disease," said first author Bin Chen, assistant professor with the Institute for Computational Health Sciences (ICHS) and the Department of Pediatrics at UCSF.

    In a paper published online on July 12, 2017, in Nature Communications, the UCSF team used the method to identify four drugs with cancer-fighting potential, demonstrating that one of them--an FDA-approved drug called pyrvinium pamoate, which is used to treat pinworms--could shrink hepatocellular carcinoma, a type of liver cancer, in mice. This cancer, which is associated with underlying liver disease and cirrhosis, is the second-largest cause of cancer deaths around the world--with a very high incidence in China--yet it has no effective treatment.

    The researchers first looked in The Cancer Genome Atlas (TCGA), a comprehensive map of genomic changes in nearly three dozen types of cancer that contains more than two petabytes of data, and compared the gene expression signatures in 14 different cancers to the gene expression signatures for normal tissues that were adjacent to these tumors. This enabled them to see which genes were up- or down-regulated in the cancerous tissue, compared to the normal tissue.

    Once they knew that, they were able to search in another open-access database, called the Library of Integrated Network-based Cellular Signatures (LINCS) L1000 dataset, to see how thousands of compounds and chemicals affected cancer cells. The researchers ranked 12,442 small molecules profiled in 71 cell lines based on their ability to reverse abnormal changes in gene expression that lead to the production of harmful proteins. These changes are common in cancers, although different tumors exhibit different patterns of abnormalities. Each of these profiles included measurements of gene expression from 978 "landmark genes" at different drug concentrations and different treatment durations.

    The researchers used a third database, ChEMBL, for data on how well biologically active chemicals killed specific types of cancer cells in the lab -- specifically for data on a drug efficacy measure known as the IC50. Finally, Chen used the Cancer Cell Line Encyclopedia to analyze and compare molecular profiles from more than 1,000 cancer cell lines.

    Their analyses revealed that four drugs were likely to be effective, including pyrvinium pamoate, which they tested against liver cancer cells that had grown into tumors in laboratory mice.

    "Since in many cancers, we already have lots of known drug efficacy data, we were able to perform large-scale analyses without running any biological experiments," Chen said.

    He and colleagues developed a ranking system, which he calls the Reverse Gene Expression Score (RGES), a predictive measure of how a given drug would reverse the gene-expression profile in a particular disease--tamping down genes that are over-expressed, and ramping up those that are weakly expressed, thus restoring gene expression to levels that more closely match normal tissue.

    After using open-access databases to determine that RGES was correlated with drug efficacy in liver cancer, breast cancer and colon cancer. Chen focused on liver cancer cell lines, but since they have not been investigated as much as breast and colon cancer cell lines, there was far less data available to study them. So, he used RGES scores for drugs and other biologically active molecules that had been tested on non-liver cancer cell types. The RGES scores were powerful enough that he could still predict which molecules might kill liver cancer cells.

    Chen's collaborators from the Asian Liver Center at Stanford University examined four candidate molecules with known mechanisms of drug action. They found that all four killed five distinct liver cancer cell lines grown in the lab. Pyrvinium pamoate was the most promising drug, shrinking liver tumors grown beneath the skin in mice.

    Cancer researchers usually target individual genetic mutations, but Chen said drugs that are targeted in this way often are less effective than anticipated and generate drug resistance. He said a broader measure such as RGES might lead to better drugs and also help researchers identify new drug targets.

    Because RGES is based on the molecular characteristics of real tumors, Chen said it also may be a better predictor of a drug's true clinical promise than high-throughput screening of large panels of drugs and other small molecules, which are based on drug activity in lab-grown cell lines.

    "As costs come down and the number of gene expression profiles in diseases continues to grow, I expect that we and others will be able to use RGES to screen for drug candidates very efficiently and cost-effectively," Chen said. "Our hope is that ultimately our computational approach can be broadly applied, not only to cancer, but also to other diseases where molecular data exist, and that it will speed up drug discovery in diseases with high unmet needs. But I'm most excited about the possibilities for applying this approach to individual patients to prescribe the best drug for each."