"We're mining the genome to see if a specific gene is involved in a disease," says Dalia Cohen, who heads up the 123-strong group. Half of her group is located at Novartis' worldwide headquarters in Basel, Switzerland, while the rest are preparing to move in March from Summit, N.J., to Cambridge, Mass., where the company is establishing its worldwide global research center.

While most big pharmas depend both on internal efforts and collaborations to identify drug targets, Novartis' early start sets it apart.

"They are doing a better job than the majority of the industry because they have taken a really disciplined approach and made it a priority to understand what's going on," says Phillips Kuhl, president of Cambridge Healthtech Institute.

Reflecting its systems biology approach to drug discovery, the group consists of five units. They include proteomics and nucleic acid groups, both in Basel (and the only pre-existing units when the functional genomics group was formed in 1998). The units focusing on model organisms and molecular biology will be relocating to Cambridge.

On target: Dalia Cohen is head of Novartis' functional genomics group, which has already developed about 25 drug targets.

The unit has grown from two Sun servers in 1998 to about 150 CPUs today, consisting of Sun servers plus a collection of desktops that includes no-name PCs. That may sound like a lot of CPUs, although it is dwarfed by Celera Genomics' recent purchase of a dozen IBM mainframes and 150 terabytes of storage from EMC.

Reluctance to bulk up on computers speaks volumes about the department's philosophy of farming out compute-intensive jobs to companies such as Celera. The decision to collaborate on research — the functional genomics group has major alliances with Celera and Compugen Ltd., as well as four small collaborations — boils down to time and money.

For instance, Reinhardt's group is working with Compugen to develop a database of all expressed human genes — the human transcriptome. "To develop the underlying algorithms would take five years, and after that I would not need the people any more," he explains. "The Compugen collaboration requires 10 people [internally]. It [also] does not make sense to invest in that amount of computing resources that have a very short life [span]."

The work that goes on inside his unit is less onerous in the computing sense. "We focus on finding members of druggable gene families," he says, "where it's applying existing algorithms to proprietary data."

One innovation in his department is the increasing use of dedicated microprocessors with built-in algorithms, such as Tera-BLAST from TimeLogic Corp. Reinhardt's department uses a couple of TimeLogic's DeCypher accelerators, each consisting of hundreds of chips. "Computational scientists use these [to speed up] the pipelines for gene annotation," he says.


Challenges Ahead
Reinhardt's unit currently faces two major challenges: first, the transition from sequencing tools to mathematical analysis, and second, creating tools that can be used throughout Novartis.

"The people we are looking for understand multivariate statistical analysis, modeling, and simulations," he says. "The trend is toward mathematical analysis of huge databases. It used to be we hired people who could do sequence analysis."

Efforts to develop tools for use among a variety of interests center on building high-level interfaces, simultaneously allowing researchers to drill deeper into data and evidence trails. "There are always compromises between different audiences, and it's hard to come up with universal use," he says. "A computational biologist looks at sequence alignment. A molecular biologist might want just high-priority genes. We're trying to present data on multiple levels."

As for operating systems (Linux, Solaris, and Microsoft NT) and hardware, Reinhardt confesses to no loyalty. "I buy whatever is best for the task," he says pragmatically. For instance, many of the group's databases of ESTs (expressed sequence tags), RNA expression profiles, and SNPs (single nucleotide polymorphisms) sit on parallel Sun servers with large memory architectures.

"I could switch tomorrow," he says. "The answer depends on price performance. If SGI (Silicon Graphics) comes along with a fantastic offer, I would go with SGI."

While he views his group as more research oriented than focused on IT service, he's glad to share or even develop algorithms for Novartis groups besides Cohen's. "I am perfectly happy to have my phone ring," he says, adding that he expects his department to grow 50 percent in the next year or two.


Form to Function
Whatever Reinhardt's group does supports the notion of multidisciplinary drug discovery within Cohen's functional genomics organization. "All the [sciences] are working in parallel to identify and prioritize disease genes," Cohen says.

Looking ahead: Paul Herrling, vice president of Novartis Pharmaceuticals corporate research, predicts that the functional genomics group will produce 30 to 50 high-quality drug targets within the next year.

Once validated, the targets are transferred to disease experts who develop drug compounds and antibodies. Priority areas include targets for osteoarthritis, chronic pain, schizophrenia, Alzheimer's disease, and diabetes. "Biologists, pharmacologists, and medicinal chemists [conduct] high-throughput screening to identify small molecules," Cohen explains. "Then the chemists will work on them."

Four to six years pass before a compound or antibody enters human testing, where it could remain for another three to four years, Herrling says. It takes two more years or so to validate and nominate a target for further investment, although Cohen says her group has already halved that time for Alzheimer's disease.

That easily adds up to 10 years or more to develop a drug. Even so, speeding up the process is not the paramount concern, says Herrling: "The key issue is the quality of the target. What [Cohen] contributes is finding higher-quality targets rather than [discovering them through] serendipity."

Moreover, speed is unpredictable in nominating targets, Herrling says. "There are some genes you find in yeast that you find in humans ... For others, you will have to use the worm." Other genes of interest lacking counterparts in lower organisms require experiments in mice, which could add another couple of years.

Disease genes are often tested in several model organisms at once, Cohen adds. "It's not a linear process," she says. "[Testing sequentially] would take too much time and is the old way to do research. [We have] changed the paradigm in the way we do all research. Whatever we do in a model organism, we will have to do in mammalian systems."

Cohen is more comfortable discussing what her organization does rather than the specific tools used to conduct research. "We look for genes and proteins that are differentially expressed in nondisease tissue," she says. "We want to know what causes the disease, not just the symptoms. For example, are there genes that can induce phenotypes that have the characteristics of arthritis in cartilage cells? This really combines genomics, high-throughput screening, cell biology, and [computational biology]."

If any doubts remain about Novartis' commitment to post-genomic science, the new $127- million, 410,000 square-foot Genomics Institute of the Novartis Research Foundation in La Jolla, Calif., where 350 researchers work, should put them to rest. (Cohen won't reveal her budget, but Novartis could spend $250 million this year on Cohen's group and the Institute alone.)

More freewheeling and independent, the Institute is nonetheless on a convergent path with Cohen's genomics group. But that doesn't mean they'll be merged, Herrling says. "Functionally they collaborate, but you have to give scientists autonomy."