Jim Rizzi, Array Biopharma’s director of computational technology, has a simple philosophy when it comes to the company’s informatics needs: “If you can buy it, buy it; but if you can’t and you need it, let’s see if we can’t come up with something.” So far, the cases where his team has opted to build rather than buy have proven successful enough that the company is now developing an in-house desktop-based cheminformatics system, in collaboration with CambridgeSoft, which will be used by the entire Array Biopharma research staff.
Off-the-shelf desktop tools, according to Rizzi, “were for everything — they were for biology, genomics — they were just great graphics tools and they were great at what they do, but we didn’t need that whole big suite of things. We said we just want the basic chemistry core tools that a chemist will sit there and obviously understand without having to use it every day.”
The long-term goal of the desktop system is to pass on to the company’s broader research staff some of the data-mining activities that are currently performed by a core team of around five computational chemists that Rizzi calls his “expert user group.” Once the company-wide system is up and running, more than 180 of the company’s scientists will have ready access to all Array Biopharma’s data, as well as some key analysis tools, in the form of an electronic notebook-based system. This will free up the expert group to develop more advanced applications, Rizzi said.
Array, based in Boulder, Colo., specializes in small-molecule drug discovery. Rizzi claimed that the company is like any other drug company in many aspects, but it does differ a bit in its attitude toward computational technology. “The majority of what makes this stuff work is the buy-in by the organization,” he said. Unlike other biopharma companies where computational scientists are often treated as “offshoot, fledgling groups,” Rizzi said that his team and the methods they have developed have been “fully bought into by the organization.”
In addition to the expert group, Rizzi oversees a team of around eight IT specialists who handle hardware and infrastructure issues, as well as a team of scientific programmers who write the code for methods developed by the computational chemists. Additionally, Rizzi said that three scientists act as “liaisons” between the computational and bench chemists, and ensure that the tools his team is developing suit the needs of the research staff.
The group has a number of in-house tools in its repertoire. One, called MASC (Multiple Active Site Corrections), is a statistical method for ranking hits generated by the Gold and FlexX docking programs. Rizzi said that one drawback with current docking methods is that the binding numbers generated by the programs “don’t always correlate with [actual] binding data.” MASC runs each molecule against several random active sites that it doesn’t bind to in order to get an average binding number that serves as a baseline. “Then we run it against our target and see how much better we’re binding relative to an average number, so what you’re doing in essence is sort of flattening out the non-specific binding.” Rizzi said that this approach has eliminated a “great percentage” of false positives generated by the docking programs.
Array’s informatics team has also developed two structure-based drug design programs, called Delve and Graph, and the company is currently “in clinical candidate-seeking mode on one of the projects that came from this approach,” Rizzi said. The company has also developed several tools for predictive toxicology using both a rule-based approach that flags certain chemical moieties to avoid as well as a QSAR approach that relies on predictive modeling to determine the probability of a molecule exhibiting toxicity.
The success of Array’s computational group is “based on how they contribute to the bottom line, which is, in our case, patents,” Rizzi said. “How do you value a modeler? Based on the number of calculations he does?” he asked. “No, we evaluate based on whether they are contributing to the projects they work on, and that usually shows up in patents.”
The desktop-based end-user system that Array is developing in collaboration with CambridgeSoft is just about halfway into what Rizzi estimated is a two-year project. So far, he said, the companies have put together a solid system for storing and retrieving the company’s data, along with some general-purpose visualization and data-mining tools. Over the next year, he said, “We’ll add more tools into it that will be proprietary once we get the main foundation.”