YouTube Facebook LinkedIn Google+ Twitter Xingrss  



By Malorye Branca
Senior Informatics Editor

April 15, 2003 | The advent of DNA chips in the late 1990s came, conveniently, just as genome-sequencing projects were gaining momentum. At last, scientists could study the expression or sequence of thousands of genes simultaneously.

 Spot the difference: Expression of 128 genes analyzed in 64 primary and 12 metastatic cancers reveals a subset of 64 genes overexpressed in metastases (red, bottom right) and some primary tumors (where the arrow points).
"Three years ago, people had the idea that studying gene expression ... would somehow open our eyes to disease pathways never seen before," says Anthony Brookes, of the Karolinska Institute. But there were problems — in chip manufacture, in training, and particularly in data analysis. "This turned part of biology into a data-rich area," says Terry Speed, at the University of California at Berkeley. "Where before you had a notebook to write your results in, now you have megabytes in the computer."

Biologists weren't used to that, and it showed. In the early days, says Nat Goodman of the Institute for Systems Biology, researchers declared gene expression as "significantly different if you saw a twofold or threefold change, without any attention to replication or statistical criteria."

Gary Churchill, a staff scientist at the Jackson Laboratory, chanced upon one of Stanford University microarray pioneer Pat Brown's groundbreaking Science papers while surfing the Web: "A lot of statisticians like me noticed and downloaded the data. I carried those [results] around on my laptop for a long time, trying to figure out what was going on."

The breakthrough came when postdoc Katie Kerr showed him a striking graph (see "Something fishy"). "With a data set like that, you need replication so you can sort out the signal from the noise," Churchill says. Kerr's graph showed that the two fluorescent dyes typically used in microarray experiments had different intensities, and that was skewing the data plot. "A lot of excellent statisticians have migrated to the field," Goodman says, "but Churchill and Speed were among the first, and pushed the hardest." Both scientists maintain useful Web sites (see www.jax.org/staff/churchill/labsite/ and stat-www.berkeley.edu/users/terry/zarray/html).

Results from commercial and homemade DNA chips have improved dramatically with experience. "The goal was to get the noise from the chip to become irrelevant," says Affymetrix President Steve Fodor. The payoff is coming, although in a slightly different way than anticipated. "Rather than a research tool that gets you to the primary cause [of a disease], it's another type of phenotype," Brookes says. In other words, chips reveal important differences, but not necessarily the reasons for those differences.

The past two years have witnessed a stream of reports dissecting gene expression "signatures" in cancer. At the Whitehead Institute Center for Genome Research, Sridhar Ramaswamy, Todd Golub, and colleagues described a signature that predicts whether tumors are likely to metastasize.

This study used data from a variety of DNA chip platforms, and several tumor types, showing how much more robust the data and the analytical tools have become (Ramaswamy, S. et al. Nature Genet. 33, 49-54: 2003).

Further enhancements are on the horizon. "If protein chips could really work, they could be very valuable," says Scott Patterson, who just joined Farmal Biomedicine from Celera Genomics. "The real future lies in the integration of data from many different sources — genotypes, proteins, metabolites, and arrays," Churchill predicts. "It is really essential to find new ways to tie them all together."

—Malorye Branca



Back to Beyond the Blueprint 


RAMASWAMY, S., ET AL, NATURE GENETICS 33. COPYRIGHT NATURE PUBLISHING GROUP, USED WITH PERMISSION



For reprints and/or copyright permission, please contact  Terry Manning, 781.972.1349 , tmanning@healthtech.com.