By Salvatore Salamone
December 15, 2003 | BOSTON -- One of the biggest challenges life science organizations face is striking a balance between industrializing their data analysis procedures and incorporating new R&D investigative tools and techniques as researchers develop them.
Participants in a panel discussion at the recent Spotfire 2003 User Conference* agreed that formal data analysis procedures and workflows are critical to company success, but also insisted researchers must have leeway to deal with new types of data on an ad hoc basis.
The real issue is learning how to distinguish short-lived, “disposable” applications from those that should be incorporated into a company’s standard workflow.
“One of the key challenges is collecting and organizing information,” said Rich Lawson, global director of lead informatics at AstraZeneca. “If the information is not available, is not well structured, you’re stuck.”
“A comfortable starting point for chemists is to use [Spotfire’s DecisionSite] as a visualization tool,” Lawson said. “They can do quick analysis on their own.”
This ad hoc analysis need not be at odds with industrialization of workflow processes. “We’ve done a lot of work in industrializing [discovery] workflow processes,” said Greg Tucker-Kellogg, associate director of informatics at Millennium Pharmaceuticals. But he noted that elements of the process often start with a researcher asking simple questions.
“We have people using DecisionSite doing ad hoc analytics,” Tucker-Kellogg said. “These are disposable apps.” But when a particular query gets used over and over again, he said, “it gets integrated into the larger workflow process.”
“It’s a series of steps,” said Andy Palmer, senior vice president of operations at Infinity Pharmaceuticals. “It might start with a scientist writing an SQL query of a data set. If this query is used often, the scientist would naturally save the query.”
This one query might also be e-mailed to others in the company who are doing similar research. If this stored query gets used a lot, it is turned into a procedure that is then incorporated into a more formal workflow. “We might wrap it in Java and make it into a Web service,” Palmer said.
He noted that companies don’t want to spend a lot of time or money developing formal analysis applications for every avenue of investigation. With these mini-applications and stored queries, “a lot more are disposable than reusable, due to the nature of the work.”
“We feel that these apps are disposable, with a half-life [often] measured in weeks and no longer than nine months,” Palmer said. “So we don’t want to spend a lot of resources developing them.”
Once the bench scientist finds a useful application, that process might be stored and formalized. Specifically, the small analytic process is captured. “We repurpose and reuse these applications,” Palmer said. He noted that these discrete applications become components that are made available throughout the company as Web services.
This approach offers a unique advantage. Traditional software development creates monolithic applications for specific tasks. “You end up with [software] silos,” Palmer said. “And there is limited reuse of software.” In contrast, with a Web services approach, he said, “you end up with composite apps.”
* Spotfire 2003 User Conference, Boston, Oct. 28-29, 2003.