Protein structure modeling has received a much-needed shot in the arm with the formation of the Joint Center for Molecular Modeling. Armed with a three-year, $2.1 million seed grant from the NIH, Adam Godzik, of the Burnham Institute for Medical Research, and Pavel Pevzner, of the University of California, San Diego, will lead the new center in developing software for enhancing protein structure modeling and prediction.
"The feeling in the field is that protein structure modeling [has been] pretty much at a standstill for the last 10 or 15 years," Godzik says. "The first paper on structure modeling was published in the 1980s, and it almost reads like a textbook example of how people do modeling now." He attributes this stagnation to a lack of accurate prediction methods to determine how protein structures might change in response to mutations.
Currently, researchers classify proteins together according to the size and location of specific parts of their structures. Sequence analysis can help investigators predict that a protein found in bacteria would be similar in structure to its human counterpart, but the ability to fully model the detailed differences between the two is limited.
Godzik and Pevzner have decided to remedy this lack of accuracy with a fresh approach to structure classification. By combining Pevzner's algorithmic know-how with the thousands of already solved protein structures, the team has set out to identify common threads in protein development. "What we've suggested is perhaps we can use some tools that Pavel has developed in graph theory and start to classify changes in protein structure empirically," Godzik says. They believe that much in the same way you can mathematically reduce the complex movements of a wind-blown skyscraper to that of a swinging pendulum, it is possible to isolate simple rules and patterns out of the dauntingly intricate variations of protein development. "We imagine that when we actually map [protein movement], it would really turn out to be one or two generalized directions," he says.
Eventually, the center aims to develop a software application that can make protein prediction and classification a whole lot simpler. "There would be a program where we would just feed it a couple of structures, it would shoot out some sort generalization of what the structure family looks like, and then predict what other proteins in this family could do," Godzik says.
But for now, the biggest challenge is showing the NIH that they're moving in the right direction. "In three years we cannot solve something which people have not been able to solve for 20 years," Godzik says. "This time we're judged on just an idea — next time we'll be judged on results."