HEIDELBERG, Germany — Around 700 self-described “systems biologists” gathered in this picturesque German city for the fifth annual International Conference on Systems Biology last week. The high attendance — a record for the annual meeting — was just one of several indicators that the field is one of biotechnology’s fastest growing disciplines, despite an apparent reluctance on the part of the pharmaceutical industry to jump on board.
This year’s conference offered proof that a solid — but largely academic — community is emerging around the use of mathematical modeling methods to help elucidate biological processes. Only around 100 attendees at the conference were from pharmaceutical or biotech firms, according to the meeting’s organizers.
The problem, according to Siegfried Neumann, a professor in the office of technology at Merck KGaA, is that most pharmaceutical researchers “think this is theoretical biology.” Neumann told BioInform that he considers himself an advocate for computational systems biology, but most of his colleagues within the pharmaceutical industry “still need a proof of principle.”
Despite recent advances in the field, pharma companies “still say that’s not enough to put major resources into systems biology as a discipline,” Neumann said. Pharma is still “suffering from the promises of functional genomics … so there is skepticism that this new wave will be the same old story.”
Nevertheless, the talks and posters at the conference offered signs that systems biology is maturing into a computational discipline of its own — one distinct from functional genomics and bioinformatics. What’s the difference? According to Olaf Wolkenhauer, a researcher in the systems biology and bioinformatics group at the University of Rostock, there’s an easy test: If you feel that you’re drowning in data, he said, you’re doing bioinformatics, which he explained as primarily a data-integration and -management task. But if you find yourself complaining that you don’t have enough data to build robust dynamic models, consider yourself a systems biologist.
Most attendees agreed that there were no scientific breakthroughs at this year’s meeting, but many said that there was important progress in a few key areas. For example, one of computational systems biology’s ongoing debates — whether to use an inference-based “top-down” modeling approach vs. a data-driven “bottom-up” modeling approach — appears to be nearing a compromise.
A number of researchers are combining both approaches now, starting out with coarse, statistically derived models that are used to help establish a hypothesis that can help determine the kind of data required to fill in the details. Using this initial model, researchers design and perform the experiment, flesh out the model with the new data, and the cycle iterates once again.
In his keynote address, Oxford University’s Dennis Noble said he uses a “middle out” approach, borrowing a term first coined by Sydney Brenner. “Jump in at the level where you have enough data,” he said. “It doesn’t matter where, provided you know how to collect more.” In Noble’s case, he began at the level of the cell to model the heart — an effort that originally began in the 1960s. With a rough cellular-scale model in hand, he said, he can “reach down and reach up” to pull in relevant information from the molecular level, as well as the tissue and organ level.
Most biological processes worth studying are on multiple scales, he noted. Cardiac rhythm, for example, is not driven by a single type of biomolecule, but by the voltage across the cell as a whole. “One goal of systems biology,” he said, “is to identify the level at which functionality appears in the system.”
Another ongoing challenge that several groups are tackling is the so-called parameter estimation problem: Reaction kinetics and other quantitative values are undetermined for even the most commonly studied biological pathways, so methods are required for estimating the values of those parameters in a way that corresponds with experimental data.
For one “simplified” model of a signal transduction network involved in apoptosis, Martin Bentele of the German Cancer Research Center explained that there were more than 50 missing parameters. Bentele said that his team developed a parameter estimation approach that demonstrates a good fit with experimental data based on a “sensitivity analysis” method that determines how the overall system reacts to perturbations of the parameters.
Julio Banga of the Spanish Council for Scientific Research described another approach to address the same problem using a stochastic global optimization method called evolution strategies, which Banga said is more effective, but “less popular” than other stochastic methods like genetic algorithms.
Other researchers are working on integrating multiple mathematical models, particularly models on different biological scales. James Hetherington of University College, London, described a middleware-based system that the UCL team is developing to support its efforts to model human liver physiology. The system is based on wrappers for models that are created in different programming languages and different scales. These models can be linked via a system of “connectors,” which are themselves controlled by an “orchestrator” that “directs traffic” using a repository of metadata that describes what each of the models is supposed to represent.
In addition, Jacky Snoep of the University of Stellenbosch presented a “modular approach” to building an in silico yeast cell by collecting multiple, component-based models in a single resource (available at http://jjj.biochem.sun.ac.za/index.html).
Where is the Payoff?
Several speakers noted that the mathematical modeling methods that lie at the core of computational systems biology are still too complicated for the average biologist to use — a drawback that will continue to delay adoption of the approach within industry and among the broader biological research community.
“Modeling is like PCR,” said Pedro Mendes of the Virginia Bioinformatics Institute. If that workhorse of molecular biology hadn’t been automated, “no one would do it,” he said. “We have to get to that point from a software point of view — that’s the promise of systems biology,” he said.
Yuri Lazebnik of Cold Spring Harbor Laboratory kicked off the conference with a provocative talk that touched on a paper he published in Cancer Cell in 2002 called “Can a Biologist Fix a Radio?” The paper described how a biologist might characterize the functional components of a radio based on whether it stops functioning after those components are removed — as opposed to studying a formal circuit map, as an engineer would do. Biologists need circuit maps of their own, Lazebnik said, but they’ll need some help deriving them — and using them.
“What I would like to see as a biologist is developers who can create cool tools, but they need to be easier to use as a standalone unit,” he said. “I hope there is someone here [at this conference] who can write some software and actually make it usable.”
In addition to usability issues, systems biology approaches have been slow to catch on in pharmaceutical companies because, as Merck’s Neumann said, “industry is now more concerned with problems around pharmaco-economics, and the deficit of new candidates. The mindset is now set in terms of defense” as opposed to new technology investments, he said.
But there are some signs that systems biology is beginning to have an impact on drug discovery. At a satellite workshop called Industrial Perspectives of Systems Biology held prior to the main conference, representatives from several pharma and biotech firms, including Bayer Technology Services, Novo Nordisk, Lundbeck, and Cellzome, shared their experiences with biological modeling and simulation.
Andraes Schuppert of Bayer Technology Services said that modeling is likely to have its first applications in identifying responders and non-responders in clinical trials, but the “primary goal is to reduce the attrition rate in the late phases of drug development” by identifying earlier in the process drugs that are likely to fail.
Indeed, a number of speakers at the conference pointed out that Merck & Co.’s recent withdrawal of Vioxx from the market due to adverse cardiovascular and cerebrovascular effects might have been preventable if better biological models were available while the drug was in early development. According to Oxford University’s Noble, the heart model he has been developing is already capable of predicting the organ-level effects of a number of small molecules.
“It’s important to look at problems that can be of use practically,” he said. “And in the case of the heart, we can achieve that … we can now reconstruct [the heart] at several scales at once, and we can do so in a way that I hope will be of great practical value.”