WASHINGTON, DC A new version of microarray data-analysis software created by researchers at the University of Maryland and the Children's National Medical Center could allow microarray users to do power analysis to determine how many chips they will need for their experiments, according to a CNMC official.
Eric Hoffman, who heads the CNMC's Research Center for Genetic Medicine, said that there was a need for an interactive statistical tool that would allow researchers to do a power analysis prior to running an experiment to see how many chips would be needed for a successful experiment.
Hoffman claimed that the tool, called Power Analysis would be added to CNMC's Hierarchical Clustering Explorer array data analysis website when version 4.0 of HCE comes out in September. He said a tool for calculating how much "power" a researcher would need in his microarray experiment had been lacking from the field.
"Power analysis in human studies is ancient history and de rigeur," he told BioArray News in an interview after his presentation at the Cambridge Healthtech Institute's Total Microarray Data Analysis and Interpretation conference, held here last week. "You have to do it before you can even get IRB approval to do your subject study."
"Power analysis in human studies is ancient history and de rigeur. You have to do it before you can even get IRB approval to do your subject study.
"In any other organism it's generally not done," he said. "They just basically run a bunch of experiments until you find what you want to find. So it's very empirically figured out. But with chips you are in this interesting space where they are increasingly used in therapeutic studies or even to monitor different outcomes or to design therapy, yet people have gotten away with not doing power analysis simply because they say that nobody's done it," he said.
Hoffman used the example of a researcher trying to look at the interfering gamma pathway using microarrays. He said that with the Power Analysis tool a researcher could determine how many arrays they would need to see all the genes associated with that pathway. If the tool recommended 100 chips, the user could then select 90 percent of the pathway and perhaps get a more suitable number of experiments.
"It would also tell you what genes you were missing in those experiments, so you could perhaps use RT-PCR instead to look at the genes that would be left out," Hoffman said.
He said that the solution Power Analysis presented was two-fold: It would help "alleviate the multiple testing problem" by giving researchers a way to abandon the empirical model, and it would also allow them to design what he called an "ethical experiment" that didn't waste human samples on experiments that wouldn't affect the outcome anyway.
"Right now it is technically unethical to do a human experiment without doing a power analysis," Hoffman said.
But if it is unethical to run human experiments without doing a power analysis, then why haven't array users been doing them? Hoffman said that he thinks that researchers have, up until this point, looked at doing a power analysis in an array experiment as impractical because they aren't just measuring one probe, but thousands. Hoffman said that the new tool dealt with the statistical reality that array users face.
"It's all based on the formulas for power analysis, but it takes the variance of measurement for each probe set and figures out what kind of P-value and what kind of fold change you want to see, what your sensitivity [is], and how many arrays you would need to see that," he said. "It just does it on the whole array as a unit so that you can say, 'With this number of probe sets I am powered sufficiently to go into this experiment and interpret the results and with this number I'm not.'"
The Spotfire Connection
According to Hoffman, the Power Analysis tool will be free and he and his partners at the University of Maryland have no plans to commercialize it at this time.
However, he said that the group does see some commercial potential for the concept, and that ideally in the future people studying humans would be obliged to run a power analysis before proceeding with their experiments.
One way the tool may enter the bioinformatics marketplace is if companies license it from the University of Maryland, he said. There is also the remote possibility that the project could be spun out.
Ben Shneiderman, a professor of computer science at UMD, and a former member of Spotfire's board of directors who created some of the informatics company's tools, was integral in designing the new HCE version 4.0 interface and the Power Analysis tool, according to Hoffman. Shneiderman and Hoffman both worked with researcher Jinwook Seo in designing the tool. Seo is affiliated with both UMD and CNMC.
Shneiderman told BioArray News this week that he didn't see HCE as a commercial tool, but that it had some strengths that set it apart from software packages on the market.
"I think it's rightly characterized as a university prototype and not a commercial tool. It would take a further effort to make a good commercial tool," Shneiderman said.
Still, he said that the new HCE package would have some resources that Spotfire lacked, such as the ability to rank data by feature.
"Spotfire was from our lab also and it became a commercial success story and remains a commercial success story. We took on a set of other concerns in developing HCE. [We] started with the hierarchical clustering technique, which is also in Spotfire, but then [added] the noteworthy piece, a rank-by-feature framework, which is not in Spotfire," Shneiderman said.
He said that the applications were already starting to find their way into the marketplace and that one company, who he didn't disclose, had already taken a license to possibly embed HCE in its software.
Hoffman agreed that in the future companies might see the tool as useful within a larger offering. "There are some company's being formed that could use this kind of thing, but it wouldn't be the focus of the company, it would just be a tool," he said.
— Justin Petrone ([email protected])