Predicting the Structure of Important Drug Receptors
Structure-prediction algorithm searches for most likely conformation
If you want to find a Tab ‘A’ that will fit into a Slot ‘B’, you’ll waste a lot of time if you don’t know the shape of the slot. For scientists trying to design new drugs, that is sometimes the precise problem: They seek a molecule that will snug itself into a nook whose shape is unknown, difficult to determine, and capable of changing as the fit is induced.
Now, a new computational tool promises to help rescue researchers from the task of fitting square pegs into undefined holes. It models the structures of the largest family of cell surface receptor proteins in the human body: G protein-coupled receptors (GPCRs). These receptors are encoded by about five percent of human genes and are the targets of about 45 percent of all modern medications. The 3D structures of most GPCRs are unknown because the molecules are extremely difficult to work with. Like all proteins residing in cell membranes, they tend to fall apart when plucked from the membrane for analysis in a laboratory. Traditional approaches such as NMR and X-ray crystallography have only yielded a single GPCR 3D structure.
To sidestep the difficulties of the experimental approach, Jeffrey Skolnick, PhD, director of the Center for the Study of Systems Biology at the Georgia Institute of Technology in Atlanta, and his research team developed a structure prediction algorithm called TASSER. It takes whatever fragmentary information is known about a protein’s structure—or can be reasonably inferred from knowledge about related proteins—and feeds it into a structure assembly algorithm that combines the data in different ways, searching for the most energetically stable configuration.
“By looking closely at structures that are similar, you should be able to enhance drug discovery by not only designing towards what you want, but away from everything else,” says Skolnick, who estimates that of the 907 GPCRs in the human genome, TASSER has produced 820 models that are likely to be correct. The work was published in PLoS Computational Biology in February 2006.
Because no one has determined the structure of these 900 proteins, an algorithm that can produce accurate predictive models should prove significant, comments Harold Scheraga, PhD, emeritus professor of chemistry and chemical biology at Cornell University.
Skolnick emphasizes that while he’s confident most of the TASSER-generated models provide new insight into the GPCRs structures, he doesn’t expect that many of the structures have been fully deciphered by this round of modeling. “What we’re trying to do as best we can, is establish the plausibility of these [models] as hypothesis generators,” he says, which should help guide drug development research away from dead ends and into productive avenues, where the tabs and slots of medication and receptor are most likely to mesh.