Analyses and comparisons of microarray platforms are often reduced to Venn diagrams to graphically illustrate concordance.
That’s not enough, according to an Affymetrix scientist speaking at the Macroresults from Microarrays conference in Boston last week.
“Venn diagrams are too limited a way to look at things,” said Tom Ryder, senior director of assay development for Affymetrix, during a panel discussion on standardization and validation at the close of the two-day conference.
Ryder said that alternate methods of data presentation should provide gradations of significance to address a need for “new analytical tools to define if information is coalescing to tell a common story.”
Margaret Cam, director of the microarray core facility for the National Institute of Diabetes and Digestive and Kidney Diseases of the NIH, has done just that. Her group is preparing to submit a cross-platform study to a peer-reviewed journal. The information, which does include Venn diagrams and rank-order methods, was originally presented in a poster at a microarray conference held in Zurich this winter.
“Being a core facility, we need to make direct comparisons across array platforms, or it will be too difficult to analyze anybody’s array data,” Cam told BioArray News. “Eventually, the industry will have to standardize to some extent. The commercial outfits will have to start thinking about standardization and find ways to help investigators make more sense of the data.”
The conference discussion was spurred by Venn diagrams com-paring data produced by the various microarray platforms, which seem typically to contain small areas of overlap. What these representations show is that microarrays are like jazz: They produce a magnificent sound, in this case, of scientific discovery, within a framework of individual instrumentalists riffing on a theme.
The industry, however, needs to metaphorically seek the standards of a symphony, with each note rigorously following a composer’s score and the baton of the conductor if it is to create consistently reproducible results.
The “telling of a common story” that Ryder referred to for microarray platforms is being drowned out by the cacophonous complexities of biology and the technology used to measure genetic expression.
At a separate session, Andrew Brooks, an assistant professor at the University of Rochester Medical Center, and the director of the Functional Genomics Center, presented results from a study, “Assessing the Sources of Variability in Microarray Data: A Multi-institutional Study of Microarray Variability and Standards.”
The study was conducted on some 4,000 microarrays from different institutions and companies, which were analyzed to assess variability.
According to Brooks, the biggest source of data variability came from processes in the individual laboratories themselves, rather than variability introduced in the manufacturing of the arrays.
“Lab-to-lab variation is the greatest source of error,” he said.
Stephen Tirrell, director of transcriptional profiling for Millennium Pharmaceuticals, which uses the Affymetrix platform implemented with four scanners and 10 fluidics stations, said six of his labs were doing T7 labeling. “For all six protocols, none overlapped.” Part of that, he said, might be due to reagent variability and a need to standardize reagents that go into the T7 labeling, including having one person make up all the master mixes.
“As a scientific community, we’ve got a long ways to go before we get our arms around this,” Tirrell said.
One of his suggestions was for researchers to take extreme effort to document processes, workflows, standard operating procedures and specifications. This is going on to some degree with the Minimum Information About a Microarray Experiment, or MIAME standard, but this standard mainly focuses on encouraging the inclusion of certain types of data, without necessarily addressing foundational aspects such as biology and technical bias.
“With MIAME, we have a point of reference, but we need to go beyond that,” said Brooks.
The pharmaceutical industry, perhaps the leading consumer of microarray chips and instrumentation, may become the impetus for standardization under the influence of the FDA, which is starting to show an interest in genomic data, recently releasing a draft document, “Multiplex Tests for Heritable DNA Markers, Mutations and Expression Patterns; Draft Guidance for Industry and FDA Reviewers.” (http://www.fda.gov/cdrh/oivd/guidance/1210.pdf). The review period will be followed by implementation, but that, perhaps, is years away.
In a narrower time frame, the National Institute of Standards and Technology is working to create a gold standard slide to calibrate microarray scanners, said Dile Holton, product manager for PerkinElmer Life and Analytic Sciences.
NIST is collaborating with scanner manufacturers and substrate makers to design a standard that will represent typical fluorescent microarray intensity values for CY3 and CY5 to allow standardized comparative measurements for sensitivity and uniformity between scanner technology. The group will meet this month to review the first prototype with the goal of making a final slide available later this year, Holton said in a microarray news group.
Perhaps until standardization becomes a reality, researchers might just need to hum the melody.
Bob Strausberg, director of the cancer genomics office of the National Cancer Institute, said he looks for microarray data to point to biological pathways and networks.
Microarrays are not quantitative; they are semi-quantitative at best, said Brooks.