Research At The Heretical Edge

“I haven’t asked it lately,” says Eric D. Perakslis, vice president of pharmaceutical R&D information technology at Johnson & Johnson. “But I don’t think cancer is afraid of next-generation sequencing.”

Perakslis is a cancer survivor who, in his spare time, volunteered for two years as the chief information officer for the new King Hussein Institute for Biotechnology & Cancer in Jordan. He may have a close enough personal relationship with the disease to ask it questions, but for Perakslis, all drug research comes down to interrogating diseases. Which information technology tool is used is ultimately of little consequence. The tools are there only to support a simultaneous interrogation by the individual researcher and the broader drug discovery enterprise.

At J&J, Perakslis has developed a means to help both the researcher and the enterprise—a knowledge management system that uses open-source software hosted on Amazon’s public cloud computing server. TranSMART, which supports genome sequencing analysis and other experimental data, was launched at J&J in 2009. More recently, the system was adopted by the Innovative Medicines Initiative (IMI), a program launched by the European Commission and the European Federation of Pharmaceutical Industries & Associations to sponsor collaborative research among major drug companies.

Perakslis is bemused that a major drug company has given him such free rein in systems development. After all, open-source software, cloud computing, and collaboration with competitors on shared data systems have, until recently, been viewed by big pharma as beyond the pale. “Insane is a good word,” he says. “But I prefer intentionally heretical. I am three times a heretic.”

But that is only when it comes to tools and how they are deployed and shared. Perakslis views himself as a traditionalist in regard to how they are used. “I have great friends and mentors who have been pharmacologists and biologists for decades,” he says. “Forty years ago when they were being trained, biology was all about intuition.” Researchers studied nature, formed hypotheses, and tested them. “Well, a lot of the intuition involved has been replaced with tons of data that I don’t understand. It’s almost like intuition has been taken out of it.”

Collecting and analyzing that data has required a steady increase in hardware and software spending, Perakslis says, which in turn has significantly contributed to the overall cost of pharmaceutical research. “We invested all this money in attractive and novel technology, thinking if we could just bring everything together, the drugs would fall out of the sky,” he says. “Well, guess what. They didn’t.”

Perakslis says he views bioinformatics less as a means of automatically converting data into drugs than as a means of guiding researchers in designing the next experiment or clinical trial—a means of supporting old-school science. He tried, in developing information technology at J&J, to limit investment in hardware and software that would, over a short period of time, become outdated.

Thus the use of cloud computing, which promises limited spending on hardware for a company willing to send data to third-party data banks, and open-source software, which offers savings for a company that views software as a nonproprietary laboratory tool. In particular, Perakslis used an open-source informatics framework called Informatics for Integrating Biology & the Bedside (i2b2), developed by Harvard Medical School and Partners HealthCare in Boston for the National Institutes of Health.

Perakslis explains that tranSMART’s adoption by IMI essentially makes the J&J system an open-source knowledge management platform based on software from NIH. “We built it, then we gave it back,” he says. “It is the very thing that NIH or the government would want to see happening with public research dollars.”

J&J avoided the usual protocols in building tranSMART, Perakslis says. “The system wasn’t built in what I think of as a software development life cycle, where you collect requirements, do an architecture phase, and test it and deploy it,” he says. “Instead we found the folks that are doing the best thinking in the organization. We asked them, ‘If you had a question to ask, what would it be?’ ” The thought process was condensed into a program that mimics the intuitive thinking of the researcher, he says.

Sandor Szalma, head of external innovation at J&J, was involved in implementing tranSMART at the company. He claims the system has fostered collaborative research within J&J, breaking down departmental barriers. Classical drug development, he explains, entails a linear statistical analysis process within a rigid organizational structure. “What we are trying to do is make the process faster, more interactive, and more iterative,” he says.

These were exactly the data system attributes IMI sought for its research consortia. The group had already contracted with the computing department at Imperial College London to develop an informatics platform for consortium members working on asthma therapies. According to Anthony Rowe, a computing professor at Imperial, J&J was a latecomer to the consortium, which is called U-Biopred, for unbiased biomarkers in prediction of respiratory disease outcomes.

“Eric saw the synergy between what the group was doing and the technology he’d developed in tranSMART,” Rowe says. “We saw that it gave us 80% of what we were looking for.” Given the consortium’s limited funding, the decision to use tranSMART was “a no-brainer,” he adds.

Although drug companies are beginning to join consortia, few have offered to contribute data warehouse technology built within their walls. But Garry Neil, vice president of J&J’s corporate office of science and technology, says doing so made sense.

Perakslis began designing the informatics system, Neil explains, at a time when researchers were focused on gaining a better understanding of the biological mechanisms of disease. They sought a better toolbox of biomarkers and a more robust clinical-evidence infrastructure. Working with regulators on a regimen for developing personalized medicine also called for better information management tools. “TranSMART, in one way or another, addressed all these challenges,” he says.

It also facilitates data sharing, should drug companies desire. “Eric and I have been working to expand data access beyond J&J,” Neil says. Every drugmaker, he notes, has compounds that can be advanced via experiments informed by precompetitive databases. IMI’s success in bringing together research consortia shows that most companies are willing to explore collaboration, provided the line between precompetitive data and proprietary intellectual property is clearly defined.

Meanwhile, Perakslis is looking for other lines to cross in bioinformatics. “You always want to increase the ... utility of the system,” he says. Artificial intelligence may be a new frontier. “For example, tranSMART isn’t smart enough to answer the question you didn’t ask it. But with a data store like tranSMART’s, we should be thinking of that,” he adds.