Pfizer's Biotherapeutics and Bioinnovation Center. Modeling Protein-Protein Interactions in Weeks, Not Years.
'Computationally intensive' is a bit of an understatement when it comes to this process. Early attempts by the group to use existing computational resources across Pfizer did not work well: the group's protein modeling required too many CPUs and too much bandwidth and the iterative nature of the research demanded faster turnaround times.
“What we’re doing is a few orders of magnitude more complicated than what had been routinely done for small molecule development,” says Jacob Glanville. “Running our jobs on shared CPU clusters was difficult as jobs sometimes wouldn’t complete for weeks.” Glanville set about designing a cluster that would meet the Rinat group's needs. He decided on a 72-car Xserve, partly because he had experience working with Xserve at previous labs, so he was comfortable using them and knew they supported all the software he would need. “The price for the performance was good and I knew that all the computational algorithms I needed were or could be compiled for the system. Plus, "the Xserve has an incredible amount of storage space." Having a powerful, functioning cluster allows the team to do the work it needs to do. "Before we had the cluster," explains Giles Day, "we weren't really doing this work." The choice of Mac hardware, he says, mostly came down to the most practical concerns—speed and price. "We chose the Apple cluster because it was cheap, and its fast," Day says. "When it comes down to it, it's about return on investment for us."
But Mac fits comfortably into lab life for more qualitative reasons as well. Day points out that he increasingly sees Mac used in scientific settings because of its Unix-based operating system. Since the beginning of his career, he says, "the birth of Linux and the adoption by Mac of Linux as its OS really changed the playing field for everybody." Rossi and Glanville both use Macbook Pros to manage their lab work and their personal computing. "In terms of laptops, I like the idea of having a Mac because it's a Unix-based operating system, and I'm used to doing things in Linux," says Rossi. "I always had two computers, Linux and Windows. Now I can have just one computer, and do everything there."
The Results
The software pipeline generates a large number of potential solutions, called conformations, which cluster together in sets of coordinates. Those coordinates can be used to reconstitute the complex, says Glanville, and the results are visualized in a program called Chimera, which will make renditions of the antibody and the antigen that are accurate within three angstroms. Glanville views these images on the screen of his 17" Macbook Pro, and shares them with the protein engineering department, which will confirm the results and begin using them as a template for engineering decisions.
Compared to traditional methods, Rossi and Glanville's pipeline is lightning-fast. "When it's successful," says Glanville, "it can speed things up so that you can avoid an extremely laborious eight-month process. Instead of spending half a million dollars, you're spending electricity for four days." Experimental confirmation of the results takes about two weeks.
Speed is a crucial benefit of computational modeling, says Glanville, but it's not the only one. "The other advantage is that you immediately get a working model of the interaction, which is invaluable for the engineering effort. If we can provide a model of the interaction early on, then they can use that to engineer much more sophisticated modifications to the behavior of the drug."
The Vision
Antibody therapeutics have the potential to treat a panoply of human conditions including chronic pain, cancer, and Alzheimer's disease—a special interest of the team at Rinat. "The uses are essentially limitless," says Glanville. "Basically any target that an antibody can reach and modulate can be a viable therapeutic."
The team envisions faster development of drugs that are increasingly targeted and well-understood in their mechanisms of action. "I would consider it a success if we were able to fundamentally alter the process of these drug development projects," says Glanville, "to make them run faster."
Knowing the epitope early in the process doesn't just speed up drug development, though; it may fundamentally alter the game, enabling researchers to develop medicines that work in concert with a patient's individual genes. "Some people may have a mutation in the protein that you are targeting," says Rossi, "and if you can provide a 3-D model of the complex, you can map exactly where this mutation is in the protein, and you can decide whether it will affect the binding of the antibody. So you can decide whether this therapy will be suitable for this patient. This is called pharmacogenomics." It's a new field, and Rinat is well positioned to take part.
Giles Day is already thinking about scaling his technology up. "Some of the things we want to do, we won't be able to use the cluster for, because we'll be waiting for months. So we're very interested in things like Amazon EC2 clouds. We'll use our cluster to design experimental approaches, and if they can't be executed on that cluster, we'll go use a cloud or anything else we can find."
