Technologyreview.com,
November 22,
2004
Virtual collaborations for sharing data and insights are
increasingly key to scientific success. When they work, that is.
If you're a researcher studying
schizophrenia, you can tap marvelous new tools such as functional magnetic
resonance imaging and positron emission tomography. You can combine data from
these devices for astonishingly powerful new views of how the brain works. What
you can't do is easily integrate data gathered by researchers outside your
group.
Enter the Biomedical Informatics
Research Network, or BIRN. Funded by the National Institutes of Health, BIRN is
a virtual collaboration project for biomedical big science. It aims to let a
given research community share its instrumentation, data, software tools, and
other resources over very high speed networks. One of the first BIRN testbeds
is for schizophrenia researchers, who will pool their images to create a
national treasure trove.
BIRN is a prime example of a
collaboratory: "an organizational entity that spans distance, supports
rich and recurring human interaction oriented to a common research area, and
provides access to data sources, artifacts and tools required to accomplish
research tasks." That's the definition of Gary Olson, professor of
human-computer interactions at the University of Michigan and his colleagues in
the Science of Collaboratories project, which is backed by the National Science
Foundation.
The term first cropped up in the
late 1980s. Olson's project now identifies more than 200 collaboratories,
reflecting the ambitious challenges of today's science, the extremely expensive
instrumentation that it often requires, and the availability of
very-high-capacity networks and computing resources. They support research that
just couldn't be done otherwise, emphasizes Mark Ellisman, director of the BIRN
coordinating center in San Diego . This trend, he says, "can't be
stopped."
Collaboratories can be on a grand
scale, as in the Human Genome Project, or the ATLAS Project at the European
Organization for Nuclear Research (better known as CERN). ATLAS coordinates
1,800 particle physicists in 34 countries.
They also can marry formerly
separate threads of research. For instance, the Space Physics and Aeronomy
Research Collaboratory, based at the University of Michigan, gives researchers
simultaneous access to both observations and predictive models, so they can
predict "space weather" (such as the geomagnetic storms that produce
aurora borealis events) and then see what actually happens.
But collaboratories often fail.
In one early attempt to assemble genetic information, for example, an
initiative crafted sophisticated software that didn't run on researchers' most
common software platform. In another debacle, funders "put up a lot of
money to study HIV/AIDS, but in the end people couldn't figure out how to work
together," Olson says. "There's always a delicate balance between
cooperation and competition."
The NSF-funded Science of
Collaboratories project is creating a software wizard that will let
collaboratory planners assess the risks by working through a few dozen
questions. The project expects to post the wizard publicly next spring. In the
meantime, here's a working recipe for a successful collaboratory.
1. Make sure
your research community is ready: Is it
accustomed to operating this way? Particle physicists have been working in
teams for decades, a necessity given the huge cost of their instrumentation.
(ATLAS will exploit the Large Hadron Collider, an underground particle
accelerator ring 27 kilometers around and costing at least $2 billion.)
Earthquake engineers, on the other hand, traditionally work within their own
labs. As pricetags soar for state-of-the-art lab equipment, research funders
are pushing the collaboratory concept, but "the community is having a lot
of trouble embracing this model," Olson says.
2. Tackle
big questions: Scientists may realize they need
to band together to attack truly tough problems such as genome sequencing or
HIV/AIDS. But many lead researchers "still have almost a Depression
mentality: 'You've got to hoard everything,'" says BIRN's Ellisman. That
attitude "doesn't let us get science done as quickly as it might," he
adds. "After you've published whatever you've learned about your
hypothesis, you ought to publish all your data so that other people can
hypothesize about it in different ways."
3. Get each
individual participant on board:
Individual researchers must be assured that their careers won't suffer in such
broad-scale efforts. The Alliance for Cellular Signaling, a large scale project
studying the extraordinarily complex biochemical pathways in which cells
interact, tackles this by treating data contributions as publications. There
are similar concerns for the talent you need to get onboard to build the
technical infrastructure. "In a computer science department, if what
you're doing has practical applications, you've fallen from grace," says
Ellisman.
4. Gear up
for major technical challenges: Megaprojects
such as BIRN's may juggle dozens of institutions and petabytes of data over a
decade or more. They also face unique challenges. For instance, the scanners
gathering that schizophrenia data may each come with their own characteristic
idiosyncrasies, so researchers must track which scanner produced a given image,
and try to find ways to correlate images taken by all those scanners. Even in
less ambitious collaboratories, researchers also must be comfortable with
collaboration tools that are highly customized or simply new to them. "Not
everyone has the same experience with these technologies, which can be pretty
daunting," Olson says. "A lot of the tools are a little clumsy and
need a little local support. High-paid scientists just don't have the patience
to deal with something that isn't working."
5. Put
enough resources into project management: Researchers
tend to resist spending money that doesn't go directly into science. But these
complicated projects can benefit from dedicated managers with suitable training
and experience.
6. Talk the
same talk: The InterMed project, which has
standardized clinical guidelines across medical disciplines and settings,
required a huge amount of work to establish a common vocabulary Olson says. If
participants didn't agree, say, on what "patient distress" might mean
for a heart attack victim, their procedures for dealing with such cases could
not be fully spelled out and aligned with each other.
7. Hold your
course: You need plenty of patience among
the players, especially the funders. A project might take four years to hammer
out data access issues, and then run a decade or more. You need visionary
planning and stable management to stick it out.
"Collaboration is hard in
general, whether you're doing it online or not," Olson emphasizes. And it
needs the social glue of good relations among participants. No matter how fancy
your software, he adds, "the best way to start building a personal
relationship with your colleagues is face-to-face."