- Title: Assistant Professor of Computational Biology, Fred Hutchinson Cancer Research Center and the University of Washington
- Education: PhD, MIT, 2001
- Recommended by: David Baker
As a mathematician who works on protein modeling software, Phil Bradley, an assistant professor of computational biology at the Fred Hutchinson Cancer Research Center and at the University of Washington, has had a lot to learn making the transition from strictly numbers to molecular structures, conformational energies, and protein-ligand interactions. At the University of Washington, Bradley helped David Baker develop the Rosetta software, a modeling tool used to predict and design protein three-dimensional structures, as well as protein-protein and protein-ligand interactions. At the moment, his lab’s focus is even narrower: modeling the interactions and conformations of transcription factor binding sites. “We’re really interested in the question, can you model that interaction between the protein and the DNA in such a way that you can figure out which sites that protein is going to want to bind to?”
Typically, Bradley uses molecular modeling techniques to move from a predicted three-dimensional representation of the protein to a most likely conformation, one that rests in the best possible energy state. According to Bradley, they “sort of jiggle the protein a little bit and evaluate the energy — and if it gets better, if that looks like a more favorable potential conformation, then we move to that one,” he says. “Exploring the conformational space is half the equation, and the other half is evaluating the energies using a physical chemistry-based model.” Presently, he and his colleagues are looking at several different families of transcription factors, including zinc fingers and different bacterial transcription factors, as it might be easier to move from model to actual binding sites in bacteria, Bradley says.
Bradley got his PhD in applied math at MIT, but it wasn’t until late into his program, he says, that he changed the course of his thesis away from pure mathematics. As a postdoc in David Baker’s lab he did structural modeling work on protein folding, but since joining the Hutch in July of 2007, he’s focused mostly on “trying to understand how proteins interact with other molecules.”
As for future predictions, Bradley believes that one of the field’s most serious limitations is a lack of computer processing power. Improving that, through projects such as Berkeley’s Boinc distributed computing effort, would free him up to discover more structures and interactions faster. It takes time and money to evaluate so many different possible conformations of these models, a process that could also be improved by simply writing smarter algorithms. “Clearly there’s no doubt that we could be sampling these conformations in more intelligent ways and reusing information throughout the sampling process in more intelligent ways,” he says.
Bradley also sees function as looming large. Not only are protein structures important, but their interactions with other molecules reveal much about their function, which, at the end of the day, is of most clinical value. “What I’d like to see is the field of structural modeling move to the point where, in addition to computing these structures, we can also compute interactions,” he says. “I’m not really optimistic in the very short run, but I think that that’s the goal that I’m working towards.”
Publications of note
In a paper published in Nature in October 2007 (“High-resolution structure prediction and the crystallographic phase problem”), Bradley worked with a team including Baker that used structural modeling techniques to increase the accuracy of predicting crystallographic structures of proteins.
And the Nobel goes to...
For the Nobel, Bradley’s eye is on the clinical relevance of all this data, or making functional predictions that have relevance to human disease. If he won the prize, he says, he hopes it would be for “predicting all the regulatory interactions in the human genome.”