Integration and infrastructure were the informatics buzzwords that bounced through the halls at the IBC Drug Discovery Technology conference last week in Boston.
While neither concept is shockingly novel, the issue of how exactly to build an architecture to connect the stuffed silos of valuable drug discovery data and funnel this down into usable knowledge still seems to be keeping pharma’s informatics leaders awake at night.
In last Monday’s session, “Information Technology for Drug Discovery and the Life Sciences,” informatics executives from Eli Lilly, Bristol-Myers Squibb, Pfizer, Wyeth, and others discussed approaches to the challenges of building an integrated informatics infrastructure at their companies. Following are highlights of several talks.
Building the Lilly
Sangtae Kim, vice president and information officer at Eli Lilly, opened the session by confronting that thousand-pound pink elephant standing in the middle of the room: the fact that pharma productivity has decreased despite a dramatic increase in IT spending.
Instead of making excuses for this disparity, Kim proposed a solution in which discovery informatics needs to shift from data that shows mere correlation to that which has predictive value. Part of this involves a departure from the forward linear march of traditional discovery efforts. “Imagine a world where the information flows in both directions,” he said. By using downstream information effectively, scientists could figure out much more quickly whether a project will work. However, “currently the industry is not equipped to deal clearly with the flow reversal,” because scientists are not necessarily rewarded for how often they kill projects, he said.
To reinvent the process of information flow in drug discovery, Kim posited that the principles of systems biology need to be applied to IT. “The challenge with IT is to move from a reductionist to a systems view,” he said. This shift requires a tremendous investment in architecture, which is exactly what Eli Lilly is doing: “It’s not just about the budget but a willingness to place some of the very best people in architecture,” he said. “It will have a lot to do with driving R&D productivity at Eli Lilly in the future.”
BMS Has a Smart Idea
Following this architectural call to arms, informatics executives at Bristol-Myers Squibb Pharmaceutical Research Institute discussed how the company has been working over the past four years to catch up in the informatics game. The approach the company took, said vice president of informatics Shawn Ramer, was to build the infrastructure around the business and scientific objectives rather than just letting the technology drive the operation.
This approach boiled down to a platform called Structure Modeling Analysis Research Tool, or SMART Idea, that provided a web-based front end platform to draw together the company’s cacophonous data sources. The initial version didn’t work too well: The data standardization and integration did not go deep enough, and the platform had to support 1,000 users, so the company needed “a more robust architecture” than provided by the web base, said Deborah Loughney, the director of computer-assisted drug design who spearheaded the project. So a team of 70 people from six locations and numerous departments came together and developed standard data formats and built a single data repository that could be interrogated. They topped this effort off on the front end with Tripos’ analytical tools, which allowed users to more meaningfully interrogate the data.
While this initial cheminformatics system is up and running, the company now plans to add two other repositories, one with bioinformatics data, and one with clinical data — and eventually add screening, chemistry, biology, ADMET, modeling, competitive intelligence, and literature data repositories, which could all be funneled into one central data repository.
Wyeth Gets around ‘Build or Buy’ Dilemma
Peter Smith, director of discovery research applications at Wyeth Research, described how he unhooked himself from the horns of the build-or-buy conundrum. “I’m not going to buy big applications,” he explained. “I’m going to buy components.” Smith’s group built these components into an open source network standardized around Java. Although this approach is not new, “the difference now is that advances in technology have made this approach independent of vendor, language, and hardware,” Smith said.
Recently, Wyeth applied this approach to bioinformatics, in developing Biobench, a distributed computed technology assembled from reusable components that is protocol-driven, designed to handle project team data, and easy and fast to use, Smith said.
Biobench is still in pilot stage, but users are already clamoring to add chemistry capabilities to it, to make a discovery workbench. To fold in this new layer, Wyeth plans to “mix and match best-of-breed components,” Smith said. The company is planning a combination of data federation and a warehouse mart, but the complete discovery workbench is still two to three years away.
For Pfizer, which is now in the process of acquiring Pharmacia, the informatics team isn’t sweating the integration process: The team has gotten plenty of practice over the past 18 months making the data from the Parke-Davis acquisition compatible with that from Pfizer proper. Key to this effort was developing a consistent ontology for naming different pieces of information. The company did not always resort to introducing new systems at the user level. In some cases, the local “culture” of informatics tools could be preserved at one layer, while integrated into the overall system at another layer.
Part of this effort to preserve local cultures while assembling an empire comes from the company’s informatics philosophy of making the software as easy to use as possible. A favorite among the company’s informatics pros is the book Don’t Make Me Think by Steve Krug, which preaches the virtues of making the user interface straightforward and intuitive. Pfizer went a step further, after it realized that scientists do not have time to be trained. The front-end software involves “viewlets” — simple help sheets that allow users to learn as they go — as well as a home page help tool. The company has also developed user help web sites, where the scientists can ask questions about the software — a functionality they will doubtless need as the company goes through yet another layer of informatics integration.