Fresh Start 2002: Roche's New Scientific Method

Learning to Swim in a Deluge of Data

In the genomics explosion, think of the GeneChip as the detonator. To the unaided eye, it is merely a carefully mounted piece of darkened glass, barely bigger than your thumbnail. Look closely, though, and you can see countless tiny markings on that glass. Each mark represents the essence of a human gene -- assembled one amino acid at a time on to the glass. All told, there may be as many as 12,000 different genes on a single chip.

Drug companies have been showering the GeneChip's inventor, Affymetrix Inc., with big orders since the chip first hit the market in a meaningful way in 1996, and it's easy to see why. Run the right experiments, and the GeneChip will light up the specific genes that are activated in a medically interesting tissue sample. Suddenly, hundreds of brilliant white and blue dots burst forth against the chip's dark background, like the Arizona desert sky at night. It is an awesome sight. Each time a chip lights up, you behold a glimpse of which genes might be markers for disease. Yet for all of the ingenuity involved in making the GeneChip, it has required cleverness on Roche's part to use the chips effectively within a big organization.

Take something as basic as computer capacity. Each sample run on a GeneChip set generates 60 million bytes of raw data. Analyze that data a bit, and you need another 180 million bytes of computer storage. Run 1,000 GeneChip experiments a year, which Roche did in both 1999 and 2000, and pretty soon you run the risk of collapsing your data systems. "Every six months, the IT guys would come to us and say, 'You've used up all of your storage,' " recalls Jiayi Ding, a Roche scientist. Some of those encounters were outright testy. At one point in early 1999, Roche's computer-services experts pointed out that they were supposed to support 300 researchers in Nutley -- and that the 10 people working on GeneChips were hogging 90% of the company's total computer capacity.

Dealing with the computers has been the easy part. A bigger challenge involves shunting all of that information into scientists' brains -- and making sure that they can process it well. "We used to look at several data points for each experiment," says Louis Renzetti, senior director of discovery pharmacology. "Now there are dozens and dozens." Simply dump all of that data on a scientist's desk, and one of two tragicomic things will happen: Either the scientist will want to pursue every promising lead and will end up like a frazzled amusement-park visitor, or the scientists will refuse to touch the report at all, for fear that she will never be able to make sense of it.

It has taken a while to find the right approach, says James Rosinski, one of Roche's experts in the new field of bioinformatics, which covers the management of genomic data. The key, he says, is for biologists and statisticians to start talking early about how to use data from a GeneChip experiment. "It's iterative," he explains. "We can't just take a one-shot approach and tell the biologists what they ought to be interested in. We have to interact." That way, researchers from different disciplines can gradually make sense of a giant stack of printouts, turning it into a shortlist of the best prospects.

Sometimes the smartest thing is knowing when not to run a GeneChip analysis. "Back when we were working on one gene, we could do fishing experiments," recalls a Roche researcher. "When you get data from 12,000 genes, you need to be careful about setting up experiments. You'll get data. If it isn't useful, you can waste a lot of time looking at it."

Turbulent Times Demand Flexible Teams

Most of the time, cancer researcher Barry Goggin does his best work either at his desk or inside one of Roche's laboratories. But in the summer of 1999, he noticed a switch in his work practices. "We started having corridor meetings," he recalls. "We would just start talking. It might be a couple of us from oncology and then a person from genomics. Back then, there wasn't any official way for us to work together. So we'd just get together in the corridors and design all sorts of small projects."

A bureaucratic boss might have told the researchers to stop gossiping and get back to the lab. But Roche's head of preclinical research at the Nutley headquarters, Lee Babiss, liked what he saw. In fact, he wanted to cut across traditional departmental boundaries and create interdisciplinary teams. That way, Roche could harness the energy of those corridor meetings and try to direct it toward breakthrough science.

In October 1999, Roche formed the Genomics Oncology (GO) team -- an ad hoc combination of seven Nutley researchers from the genetics and oncology departments. It was one of the most eclectic teams Roche had ever assembled. Team members had backgrounds in areas from immunology to statistics. They were born in countries as far from each other as China and Germany. But they were united by a desire to identify new targets for cancer drugs using the GeneChip and other genomics tools.